International Journal of Computing Sciences Research (ISSN print: 2546-0552; ISSN online: 2546-115X)

Advance online publication

doi: 10.25147/ijcsr.2017.001.1.110
https://stepacademic.net

Long paper
Development of Crime Reporting System to Identify
Patterns of Crime in Laguna
Mia V. Villarica
College of Computer Studies, Laguna State Polytechnic University, Philippines
mia.villarica@lspu.edu.ph
(corresponding author)

Francis F. Balahadia
College of Computer Studies, Laguna State Polytechnic University, Philippines
francis.balahadia@lspu.edu.ph

Jonardo R. Asor
College of Computer Studies, Laguna State Polytechnic University, Philippines
jonardo.asor@lspu.edu.ph

Gene Marck B. Catedrilla

College of Computer Studies, Laguna State Polytechnic University, Philippines
genemarck.catedrilla@lspu.edu.ph

Date received: May 5, 2022

Date received in revised form: July 6, 2022; July 10, 2022
Date accepted: July 10, 2022

Recommended citation:

Villarica, M. V., Balahadia, F. F., Asor, J. R., & Catedrilla, G. M. B. (2022). Development
of a crime reporting system to identify patterns of crime in Laguna. International
Journal of Computing Sciences Research. Advance online publication. doi:
10.25147/ijcsr.2017.001.1.110

ABSTRACT

Purpose – This study developed an online crime reporting system that uses artificial
intelligence to analyze crime incident reports to provide up-to-date crime statistics, map
crime hot locations, and manage dynamic databases.

Method – The knowledge discovery process in databases (KDD) was utilized for the model
development. Scrum, an agile development technique, proved helpful in the iterative and
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0),
which permits unrestricted use, distribution, and reproduction in any medium provided the original work is properly credited.
gradual development of the system. In addition, several ICT technologies were applied,
such as geo-mapping and pattern analysis utilizing the data mining technique. The user
evaluation tool was composed of Technology Acceptance Model for the criteria and
ISO/IEC 25010 software metrics for the sub-criteria.

Results – Based on the patterns generated from the criminal records data set, the
researchers used machine learning in a prediction model generated using the Decision Tree
algorithm, revealing several important insights about the incidences of non-index crimes in
Laguna. The findings suggest that date, time, and location factors are the best predictors
of crime occurrences. Moreover, the researchers agree with the respondents' comments
and suggestions that the crime map should include a variety of graphical representations
such as a table ranking the crime rates from highest to lowest and a pie graph showing the
comparable data of analytics crime per town and cities to make the system more
interesting to any type of user. This is true for the crime analysis website's crime map for
public access user review, which ultimately came out to be acceptable.

Conclusion – The online crime reporting system provides various functions and features for
various users. This can be used to raise people's awareness regarding dangerous locations
and help agencies predict future crime in a specific location within a particular time.

Recommendations – PNP-Laguna would gain the most from the project, it is recommended
that the PNP-Laguna, municipal police stations, and the LSPU work efficiently together to
continue expanding the crime analysis website. The website offers analytics for decision-
making support in addition to covering crime management information systems across
police agencies for decentralization. Continued experimentation on index and non-index
crime datasets to develop intelligent systems capable of forecasting must be explored, as
the index crime dataset demonstrated a highly good result in the project.

Research Implications – The academe and local law enforcement agencies must
collaboratively develop strategies that reinforce the importance of community
engagement. The system provides public access through geo-mapping that supports the
community, or locally based crime prevention, instead of targeting individuals, targets
areas where the risks of becoming involved in crime or being victimized are high.

Keywords – crime incidents, crime report, crime patterns, Laguna, Decision Tree algorithm,
KDD

INTRODUCTION

Today, technology has permeated every facet of our culture. It has become a vital part
of our lives due to the numerous benefits it provides—education, medicine, transportation,
and, most crucially, law enforcement all benefits from technological advancements. Even
a brief inspection of the evolution of our efforts to prevent crime demonstrates that
technology – or more precisely, technological innovation – has been the driving force
behind the reform of crime prevention and crime control strategies, both by private
citizens and concerned groups and by traditional police agencies (Reichert, 2001; Chan,
2001; Harris, 2007).

In recent years, the Philippines has seen an increase in crime. According to the
Philippines, the index Crime is the primary delinquent record in the Philippines National
Police (PNP) due to unplanned urbanization, migration, a high rate of unemployment, and
poverty. As for the PNP's techniques, they take immediate action when a concerned citizen
reports and accuses someone of engaging in criminal activity or being suspicious of
anything, and they also patrol areas where there is a risk for crime. The lack of
preparedness and awareness of public safety is one of the factors in the Government issue.
Despite these rising rates, this project places a premium on crime prevention. It could be
reducing criminality in a specific area's percentage or preventing risk before it occurs using
data mining technology. The research analysis presented in the 14th National Convention
on Statistics in 2019 organized by Philippine Statistics Agency (Baculinao & Ceballos, 2019)
revealed that Cavite, Laguna, Batangas, Rizal, and Quezon (CALABARZON) should focus on
policies that will help prevent cases of crimes and improve crime solution efficiency.

As the new century accelerates technological advancements, real-time reporting of

crime situations is now feasible via mobile and web applications. Numerous tools and
techniques are used to address crime-related issues but implementing a comprehensive
crime information system that manages crime report databases, criminal records retrieval,
and the integration of crime prediction models with spatiotemporal patterns of incidents
for forecasting is infrequent. Roth et al. (2013) define crime analysis as the systematic
collection, preparation, evaluation, and dissemination of data about criminal activities to
aid law enforcement in their goal. It is a collection of systematic and analytical processes
for determining crime patterns across time (Chauhan & Sehgal, 2017). The criminal
investigation is critical for finding crime hotspots. The term is most frequently used to refer
to the entire process of obtaining valuable knowledge from data, of which data mining is a
subset. Combining data mining and crime analysis makes it possible to assess whether a
region is at risk of crime based on the datasets and attributes provided, including the crime
type, location of the crime, and date/time of the incident. There is a wealth of data relating
to crime in police station records; information about the specific crime or essential
information relating to the crime should be extracted (Hassani et al., 2016). It enables the
department to carry out various activities, including patrol deployment, special operations,
tactical units, investigations, planning and research, crime prevention, and administrative
services.

A subset of data analysis called predictive analytics is being utilized by policymakers to

help them make better decisions. By utilizing a machine learning algorithm, this
methodology for crime investigation becomes well-known and advantageous (Ippolito &
Lozano, 2020). Machine learning algorithms such as k-NN, Naïve Bayes, and decision trees
were used in classifying crime in a small amount of data (Wibowo & Oesman, 2019). In an
experiment, it is found that the decision tree was a better performing machine learning
compared to Naïve Bayes after gaining a much higher accuracy, precision, and recall (Iqbal
et al., 2013). It is proven to give a high accuracy; thus, it can still be improved by integrating
ensemble methods or application of different feature selections (Aldossari et al., 2020).
The decision tree is a reliable predicting algorithm when integrated into computer systems
(Ahishakiye et al., 2017). Furthermore, other classification techniques were used to improve
the performance of decision trees like regression (Sapin et al., 2021).

Criminal analysis and investigation examine and determine the relationship between
crime and offenders (Chen, et. al., 2015). There is a wealth of data relating to crime in police
station records; information about the specific crime or essential information relating to
the crime should be extracted.

The research study aimed to design an online crime management information system
that, using predictive analytics, may contribute to reducing Laguna's rising crime rate.
Primarily, the research must accomplish the following objectives: a.) Create a crime
prediction model that analyzes index criminal episodes using various attributes from the
acquired datasets. b.) Develop a web-based crime information system that handles record
management, storage, and retrieval for police stations and administrators covering
geographical information and forecasting systems c.) Evaluate the public user view of
dynamic Spatiotemporal patterns of criminal activities in the Laguna premises.

LITERATURE REVIEW

Traditional data mining techniques such as association analysis, classification, cluster

analysis, and outlier analysis identify patterns in structured data (Han & Kamber, 2006).
Classification identifies common characteristics among various criminal entities and groups
them into predefined categories. Classification has been widely used to forecast crime
patterns, and it can significantly reduce the time required to identify criminal organizations.
The technique, however, requires a predetermined classification strategy (Sukanya,
Kalaikumaran & Karthik, 2012). Additionally, classification requires complete training and
testing data, as a significant degree of missing data reduces prediction accuracy.

In recent times, many researchers have conducted experiments to predict crimes using
various machine learning methods and particular inputs. For crime prediction, k-NN, Naïve
Bayes, and Decision trees are listed as high-performing algorithms specifically in accuracy.
k-Nearest Neighbor (k-NN) is a data classification technique based on the proximity of the
data. k-NN makes an explicit prediction on the testing data based on a comparison of k-
Nearest Neighbor. To calculate the proximity of data, the Euclidean Distance formula can
be used (Thongsatapornwatana, 2016). Bayes is one of the classification methods that can
predict the probability of membership in a class. The value of a class in the Naïve Bayes
method is independent, which independent of other attributes. It is a simple learning
algorithm that utilizes Bayes' rule together with a strong assumption that the attributes
are conditionally independent given the class (Webb, Keogh & Miikkulainen, 2010). A
decision tree is a tree that exists in the analysis of problem-solving and alternative solutions
mapping which can be taken from the problem. A decision tree can also be called one of
the most popular classification algorithms since it is easy to interpret. Decision trees work
well with scenarios that produce discrete values. The fundamental advantage of employing
a decision tree is that it can be used to analyze and depict complicated decision-making
processes in a way that makes them simpler and easier to understand. Typically, decision
trees are used to gather the data that informs decisions. A node root (beginning point) is
used by the user to start the decision tree. The user solves the leaf nodes under the tree
algorithm choice based on this node root. A decision tree with each branch displaying
potential decision scenarios and their outcomes is the product of composing the node root
and leaf node (Agrawal & Agrawal, 2015).

Abdulrahman and Abedalkhader (2017) compared two types of classifications: the K-NN
classifier and the Naïve Bayes classifier. The research used dataset contains incidents
derived from the SFPD Crime Incident Reporting system. A total of (878,050) crimes are
provided in the training data and the test data for which the categories will be predicted
are of the same number data ranges from 1/1/2003 to 5/13/2015. The training set and test set
rotate every week. Python was used to explore training data, make regression analysis, and
predict categories for test data, to get the best correlation between the features (Date,
Pd-District, Address, Day of Week, Description, Resolution, X, and Y) and the target value
(Category of Crime). Several trials of different Regression methods were used on the
training data by splitting it into two sets; training and validation, both validation and cross-
validation were conducted, and the method with the least Log loss was applied to predict
the results for the test data. In this research, the two main algorithms were used: k-NN and
Naïve Bayes. In the K-NN approach, they applied Uniform and Inverse versions of the
technique. For the Naïve Bayes approach, they used three different types; Bernoulli,
Gaussian, and Multinomial. They concluded that compared to Naïve Bayes, K-NN has some
benefits and drawbacks. As opposed to Naïve Bayes, which can only have linear, elliptic, or
parabolic decision borders, k-NN 's decision boundary can assume any shape. Additionally,
Naïve Bayes is not effective with correlated features; if correlation rather than marginal
distributions serves as the defining feature of classification, NB is not the best option. The
lack of an attribute can also mislead Naïve Bayes. One of the drawbacks is that k-NN simply
considers distance and does not consider the most crucial characteristics. As a non-
parametric model, k-NN cannot reveal any correlations between the distribution of
attributes and classes, making it less interpretable than NB. Naïve Bayes simply ignores the
attribute of missing data, while k-NN does not appropriately handle missing data. In k-NN,
K's value must be adjusted, and an ideal value must be chosen. Another drawback is that
k-NN takes longer to process predictions when processing big volumes of data; the speed
difference is noticeable.

However, in the study of Sathyadevan and Gangadharan (2014) which created a system
that can predict regions that have a high probability of crime occurrence using the Naïve
Bayes algorithm they create a model by training crime data related to vandalism, murder,
robbery, burglary, sex abuse, gang rape, arson, armed robbery, highway robbery,
snatching, etc. extracting data from different web sites like news sites, blogs, social media,
RSS feeds, etc. that gave them 90% accuracy. Another advantage of Naïve Bayes is that it
works well for a small amount of training to calculate the classification parameters.

Agrawal and Agrawal (2015) stated in their study that when decision tree and bayesian
techniques are compared, though the accuracy of a decision tree is a far better
computational time of Bayesian network is low. Hence, when the data set is very large it
will be efficient to use NB models. In addition to this, Iqbal et. al. (2013) compared the Naïve
Bayesian and Decision Tree for predicting 'Crime Category' for different states in the USA.
Using dataset acquired from UCI machine learning repository website which contains a
total number of 128 attributes and 1994 instances. The results from the experiment showed
that the Decision Tree algorithm outperformed the Naïve Bayesian algorithm and achieved
83.9519% Accuracy in predicting the 'Crime Category' for different states of the USA.

Wibowo and Oesman (2020) tested the various algorithm by implementing three
methods, such as k-NN, Naïve Bayes, and Decision Tree, to compare the accuracy. The study
was conducted on a data set with a total number of 1,735 crimes data for 3 years, which
consisted of 15 attributes, such as Day, Season, Time, TKP, District, Gender of the Victim,
Occupation of the Victim, Age, Residence, Sex of the Offender, Occupation of the Offender,
Age of the Perpetrators, the Residence, number of the perpetrators, and types of crimes
which were classified into 3 classes, such as theft, fraud, and embezzlement. Based on the
three algorithms, the accuracy of k-NN with k = 5 was 57.88 percent, with k = 10 was 59.49
percent, with k = 15 is 59.38 percent, with k = 20 was 60.18 percent, and with k = 25 was
61.57 percent. Meanwhile, for the Naïve Bayes algorithm, the accuracy reached 65.59
percent, and the Decision Tree algorithm reached 60.23 percent. In conclusion, the
algorithm with the highest accuracy was owned by Naïve Bayes.

The crime pattern and prediction literature examine the relationship between crime and
a variety of characteristics, resulting in the development of methodologies for crime
forecasting. Most works are devoted to predicting hotspots, and locations of different
geographic areas with a high chance of crime. Thematic Mapping is one of the approaches;
the map is divided into boundary areas, and offenses are represented as points on the map.
After that, the points can be aggregated into geographic unit areas and colored according
to the number of crimes committed (Williamson et al., 2001). This technique quickly
identifies locations with a high crime rate and allows for an additional examination of the
issue by "zooming in" on specific areas.

As a result of various literature reviews and technology, the researchers recognize the
value of extracting meaningful information from previously known (historical) crimes and
criminals' relationship records to discover the facts surrounding the committed crime and
minimize future crime risks and incidents. The information above may benefit the police
authorities in Laguna as they work to reduce criminal activity through resource allocation
and proactive crime countermeasures.

MATERIALS AND METHODS

The research undertaking is aligned with the following clusters of disciplines: web
development and machine learning, which deal with Crime Incidents in Laguna. To ensure
the project's success, various data collection, analysis, and programming tools and
techniques are used. The data collection methods used were interview and document
analysis. While interviewing important officials, information desk/center officers, and
police inspectors used to provide a thorough understanding of the system, observation
supports the interview crime reports, and its variables have helped develop the system.

Crime pattern analysis tools have been found to provide valuable insights and
knowledge about crime incidence, which can create effective crime prevention operations,
particularly in known crime hotspots. Additionally, the current study project aims to assist
the police authorities in Laguna in increasing the transmission of crime reports and
knowledge throughout the province. The typical knowledge discovery process in
databases (KDD) was used to accomplish the goals. KDD is a process for automatically
analyzing and modeling huge data repositories through exploratory analysis and modeling.
KDD is the systematic discovery of valid, unique, valuable, and visible patterns in vast and
complex data sets.

Model Development

Data Collection

Data from police records were obtained with permission from the provincial
headquarters of the PNP in Laguna province. The records include details for every crime
responded – index and non-index from 2014 to 2017 (n=7,260) such as the date and time of
the crime incident, address/location of a crime, type of crime, alert level, and status of the
report. This study focuses on index crimes, defined by the Philippine National Police (PNP),
involving crimes against persons such as murder, homicide, physical injury, and rape, and
crimes against property such as robbery, theft, carnapping/carjacking, and cattle rustling.
Target data and Preprocessed data as an output.

Data Preprocessing

Data transformation is a stage in the preparation of data, the process of extracting

usable information from data, and entails evaluating and maybe interpreting patterns to
determine what constitutes knowledge. Data cleaning was made to the crime data set to
make it more analyzable. The dataset used in the experiment contains various items Blotter
Entry #, DTG, Place, Law Violated, Victim’s Data (Name, Age, Sex, Nat), Suspect’s Data
(Name, Age, Sex, Nat), Use Of (Drug, Alcohol, Unknown, Firearms), Remarks, Status,
Actions Taken (Barangay, Prosecutor, Court) as reflected on Table 1. The experiment
focuses exclusively on factors associated with index crime data from 2014 to 2017. The
quality of the mining process's output is directly related to the preprocessed data's quality.
It also entailed utilizing the Google Maps API to determine the geographical coordinates
associated with each address/location in the dataset. The following properties were
generated because of the pre-processing:

Table 1. Dataset Attributes

Attributes Description Values
The exact date the crime
DATE occurred 31/12/2016
The exact time the crime
TIME occurred 12:30 AM
Barangay and City or Town BRGY SAN ANTONIO SAN PEDRO
ADDRESS the crime occurred CITY LAGUNA
MURDER, CARNAPPING, HOMICIDE,
MURDER, PHYSICAL INJURY, RAPE,
VIOLATION Type of crime occurred ROBBERY, THEFT
V_AGE Victim's Age 48
V_SEX Victim's Sex MALE
V_NATL Victim's Nationality FILIPINO
S_USE_OF Suspect's Used Weapon FIREARM
S_AGE Suspect's Used Age 35
S_SEX Suspect's Used Sex MALE
S_NATL Suspect's Nationality FILIPINO
ARRESTED, AT LARGE, DECEASED,
DETAINED, KILLED IN ACTION,
ACTIONS_TAKE The action was taken by the RELEASED, TURNED OVER TO DSWD
N officer-in-charge TURN OVER TO WCPD
FILED, NOT FILED, PLEASE BE
STATUS Status of the case INFORMED...
Special remarks on the case UNDER INVESTIGATION, FILED IN
REMARKS status COURT, FOR UPDATE, SETTLED

Crime data mining poses concerns about privacy (Kargupta, Liu, & Ryan, 2003) the
researchers ensure that specific data like victims' and suspects' data were eliminated as
agreed upon the initial agreement upon the receipts of the datasets. The researcher
categorized each incident's dates and times by year, month, and day of the week,
separating them into distinct columns, and converting time to a 24-hour format. Microsoft
Excel was used to visualize the results using graphs and charts to aid in further analysis and
interpretation. After a thorough examination, it was discovered that the dataset is missing
data for the state and number of crime variables. Two strategies were employed in the
current research endeavor to fill in the missing number of crimes associated with index
crimes. To begin, all four fields are checked for empty values. If all four attributes for a
specific record are empty, the entire record is considered irrelevant and is removed. From
a dataset of index and non-index crimes (n=7,260) to an index crime dataset (n=945).

Building and Training the Model

Choosing a data mining task, classification. Choosing the data mining algorithms
Selecting method(s) to be used for searching for patterns in the data, deciding which
models and parameters may be appropriate. Searching for patterns of interest in a
particular representational form or a set of such representations as classification rules or
trees, regression, clustering, and so forth.

Naïve Bayes which was based on Bayes theorem: P(A|B) = P(B|A) P(A)/P(B) applies a
probabilistic approach in the dataset (Chauhan & Sehgal, 2017). In this paper, the Bayes
theorem can be written as seen in Equation 1:

𝑃(𝑋|𝑦)𝑃(𝑦) Equation 1
P(y|X) =
𝑃(𝑋)
wherein X is set as the element to find in the line of subset which are the attributes of
the dataset and represented as xn The parameters are denoted by variable X. The notation
for X is as follows as reflected in Equation 2:
Equation 2
𝑋 = (𝑥1, 𝑥2, 𝑥3, … . 𝑥𝑛 )

From here x1, x2, x3,…… xn represents the attributes such as Date, Time, Address,
Violation, etc. After substituting the X, it will be expanded using the chain rule and the
following will be produced as illustrated in Equation 3:

𝑃(𝑥1, |𝑦) 𝑃(𝑥2, |𝑦) … . 𝑃(𝑥𝑛 |𝑦)𝑃(𝑦)

P(y|𝑥1, 𝑥2, 𝑥3, … . 𝑥𝑛 ) = Equation 3
𝑃(𝑥1, )𝑃(𝑥2, ) … . 𝑃(𝑥1, )

The dataset would be used to obtain all values. When all the elements in the dataset are
substituted, the denominator does not transform and remains constant; hence, it can be
omitted, and proportionality is produced as reflected in Equation 4:

𝑛
P(y|𝑥1, 𝑥2, 𝑥3, … . 𝑥𝑛 ) ∝ 𝑃(𝑦) ∏ 𝑃(𝑥𝑖 |𝑦) Equation 4
𝑖=1
 Since the class variable (y) includes more than two outcomes, the classification will be
multivariate, and y must be determined with the highest probability as shown in Equation
5:

𝑛 Equation 5
𝑦 = 𝑎𝑟𝑔𝑚𝑎𝑥𝑦 𝑃(𝑦) ∏ 𝑃(𝑥𝑖 |𝑦)
𝑖=1
The Naïve Bayes classifier is utilized in crime analysis and is compared to the K-NN
classifier (Jantan and Jamil, 2019 & Delima,2019). K-NN is a machine learning technique that
is predicated on the occurrence of similar objects nearby. As a result, those characteristics
that appear in proximity are deemed true and valuable.

Figure 1. Decision Tree

The data set was divided into two portions for validation and assessment purposes: the
training set and the testing set. The training set comprised around 80% of the data set,
whereas the testing set comprised the remaining 20%. This procedure was important to
assure the classifier model's accuracy and precision and the result is illustrated in Table 2.
Table 2. Train and Test Set
Train set Test set
Labels Number of
Number of Instances
Instances
Murder 94 24
Homicide 38 10
Physical
Injury 186 46
Rape 87 22
Robbery 94 24
Theft 190 48
Carnapping 65 16
Total 754 190

This paper also makes use of a decision tree as illustrated in Figure 1, it is well-known for
producing results that are comprehensible and legible in natural language. The tree
algorithm is represented visually as a flowchart, with external nodes representing the
conclusion and inside nodes representing properties that are deemed to be connected to
the class attributes (Ivan et al, 2017). The decision tree diagram is interpreted as conditional
statements for pattern acquisition in the dataset. For decision analysis, the combination of
all connected internal nodes will be written in an if-else expression.

Prediction Evaluation

The best classifier was picked based on the accuracy, precision, sensitivity, kappa
statistics, and specificity of each model as determined by the confusion matrix (Brownlee,
2014). The confusion matrix is a well-known technique for measuring the classification
performance of machine learning; it is represented as:

Figure 2. Confusion Matrix

Figure 2 illustrates the Confusion Matric which comprises four (4) distinct attributes: TN,
which stands for true negative, FP, which stands for false positive, FN, which stands for
false negative, and TP, which stands for true positive. In the confusion matrix, TN and TP
denote values that are predicted correctly, but FP and FN denote values that are predicted
incorrectly.

Accuracy is expressed as a percentage of properly predicted features; a score of 70% is

deemed satisfactory, indicating that the algorithm is capable of accurate prediction.
Accuracy can be calculated as shown in Equation 6:

𝑨𝒄𝒄𝒖𝒓𝒂𝒄𝒚 𝒊𝒔 (𝑻𝑵 + 𝑻𝑷) / (𝑻𝑷 + 𝑻𝑵 + 𝑭𝑷 + 𝑭𝑵) Equation 6

Precision is defined as the number of positive features recognized as positive by the

model as reflected in Equation7.
Equation 7
𝑷𝒓𝒆𝒄𝒊𝒔𝒊𝒐𝒏 𝒊𝒔 (𝑻𝑷 + 𝑭𝑷) / (𝑻𝑷 + 𝑭𝑷)

The recall parameter indicates the number of positive features successfully classified by
the model as shown in Equation 8.
Equation 8
𝑺𝒆𝒏𝒔𝒊𝒕𝒊𝒗𝒊𝒕𝒚(𝒓𝒆𝒄𝒂𝒍𝒍) = 𝑻𝑷 / (𝑭𝑵 + 𝑻𝑷)

Kappa is a measure of the algorithm's prediction accuracy, ranging from 0 to 1.0, with a
score of at least 0.50 deemed satisfactory (Ghoneim, 2019). Kappa Statistics. It computes
the conformity between two rates that classify all n items into m mutually exclusive groups.
The value of kappa is defined by Jo and Yoon in 2018.

Po − Pe
= Equation 9
1 − Pe

where Po is the relative observed agreement among rates and Pe is the imaginary
probability of chance agreement as illustrated in Equation 9.

The experiments were implemented in the Keras framework using the TensorFlow
backend. TensorFlow was utilized as the platform to establish a detection model for traffic
classification, and mainly uses the standard data set for related experiments. TensorFlow
is an open-source machine learning library developed by Google that can transfer complex
data structures to artificial intelligence neural networks for analysis. The classification
model chosen will be used to extract and visualize knowledge, as well as to generate
customized recommendations based on this knowledge. The established model's viability
can be leveraged to display fresh information via data visualization, which can then be used
for intelligent reasoning, policy recommendation, and other pertinent objective. The
Tensorflow model was generated as json_file which was used in the developed system
specifically in the crime map section using React.

System Development

This research was developed with software development methods, namely the agile
method. Agile software development is an approach to establishing software with the
collaborative efforts of self-regulating and cross-functional teams and the end users for
working on both the requirements and solutions (Collier, 2011). The values and principles
applied in Agile come from and support the Scrum framework (Larman, 2004). Changes in
user opinion about their needs and desires, and predictions that are not under the
requirements are the main principles of double recognition in the SCRUM method
(Schwaber, 2004)/ There are three core roles in the Scrum framework that work together
to produce a product as shown in Figure 3.
 Figure 3. Scrum Model

This work was performed during four project sprints over seventeen weeks. The initial
meeting with the PNP-Laguna has been the kick-start of the project. System requirements
were specified thru interview, observation, and acquisition of pertinent documents. The
main output was a working web-based system, developed, and tested using Scrum and its
best practices. In the end, a big data environment was used as a transparent way of
fulfilling the needs of crime analysts. The scrum team is composed of a scum master –
project leader, 2 programmers, and 3 designers. Regular meetings Upon the development
of this project, different ICT tools like geo-mapping, and pattern analysis using the data
mining technique.

Software Development Tools

The development of web applications has become a difficult undertaking due to the
proliferation of programming and modeling paradigms, notations, and technologies.
Integrated development environments (IDEs), relational database management systems
(RDBMSs), and other development tools. This research project used various software
development tools as indicated in Table 3.

Table 3. Software Development Tools Used

Software Purpose
ReactJS Frontend Framework
React Bootstrap React Components for Faster Development
React Leaflet Mapping Library for the map
Redux Toolkit Global State Management
Firebase Platform used for Authentication and Database
Axios Handling HTTP requests
Bcryptjs Cryptography for securing passwords
XLSX Used to parse excel files
Netlify Deploying the web application to the internet
Netlify Functions Serverless functions for Admin tasks
 One of them is the ReactJS library, one of the most contemporary web technologies. It
has been demonstrated to be the quickest rendering library. It concentrates on the display
portion of the MVC paradigm and is extensively used for developing large-scale
applications. Because it was designed for internal usage by Facebook, it has shown to be
an efficient and quick library in comparison to other technologies.

System Testing and User Evaluation

Alpha testing and user acceptance evaluation survey tools were used to check that the
system produced met its specification requirements.

Alpha Testing

At each level of the scrum, alpha testing was undertaken to ensure that all modules
operate as intended. This included running functional and non-functional tests to ensure
that the system's performance, reliability, and usability were satisfactory.

User Acceptance Evaluation

The survey tool was adapted from the Technology Acceptance Model (Davis, 1989) as
the criteria and ISO/IEC 25010 software metrics as the sub-criteria. TAM covers Quality
Factors, Perceived Ease of Use, Perceived Usefulness, Attitude Towards Using the Website,
Behavioral Intention to use, and experience. Sub-criteria comprise Reliability, Efficiency,
Integrity, Learnability, User Interaction Aesthetics, Operability, Usability, Functional
Completeness, Suitability, Attractiveness, Operability, Adaptability, Stability, and Suitability.
The research focusing human‐computer interaction (HCI) evaluation often use discrete
multipoint scales 7-point Likert Scale - Level of Acceptability, 7-point Likert Response
Scaling (1 = Totally unacceptable, 2 = Unacceptable, 3 = Slightly unacceptable, 4 = Neutral,
5 = Slightly acceptable, 6 = Acceptable, 7 = Perfectly acceptable)

The survey questionnaire underwent several steps of quality assurance to ensure its
consistency and trustworthiness in terms of the form and content of the questions. To
begin, questions were developed from a variety of survey techniques and tailored to the
study's objectives and relevance. Next, the researcher organized a team or group of five
faculty members – 4 ICT professionals and 1 development communication specialist –
tasked with reviewing and evaluating the questions' appropriateness, given that most
things were adapted from foreign circumstances. The researcher integrated all the
reviewers' suggestions, criticism, and comments into the final survey questionnaire.
Outside of the study's sample, a sample survey of 20 third-year BS Computer Science
students was undertaken.
 After the validation, 18 statements were produced and 1 open-ended question “Feel free
to write a comment if you have any suggestions regarding the website.” was developed to
examine the respondents’ recommendations and opinions to improve the website's public
access. The researchers demonstrated the website to the 45 randomly selected
respondents thru Google Meet and distributed the survey questionnaire thru Google
Forms with the website URL to acquire the user behavior and user acceptance of the
developed website.

Results and Discussion

The research aimed to design an online crime reporting management system that, using
predictive analytics, may contribute to reducing Laguna's rising crime rate. It focuses on
prediction model development, an online crime reporting management system for PNP
Laguna, and user acceptance and behavioral intention to use and visit the website
providing dynamic spatiotemporal patterns of criminal activities in the Laguna premises.
Model Evaluation Result

In this work, three (3) machine learning algorithms were used to develop the model for
predicting crime occurrences. Upon evaluating each model, the decision tree and Naïve
Bayes both showed good performance in classifying crimes.

Table 4. Classifier Performance

Classifier Accuracy Kappa Precision Recall F-score
Naïve Bayes 70.34 0.44 74.52 70.64 72.52
k-NN 64.45 0.16 72.68 633.76 67.92
Decision Tree 84.75 0.64 81.76 83.65 82.69

As the result of the experiment (Table 4) illustrates the classifier performance of Naïve
Bayes, k-NN, and Decision Tree. The decision tree showed a very promising result garnering
a total of 84.75 accuracy score and an 82.69 score for the f-measure. The model generated
from the experiment was integrated into the public access-crime map as seen in Figure 4
which represents the Geographic Information System (GIS) facility of the website which
analyzes and displays geographically referenced information specifically crime instances in
a unique location.

Developed System Result

The website was developed using Scrum methodology, inputs like functions and
features were based on the initial interview and document analysis conducted. Monthly
scrums were created and developed comprised of various such as Content Management,
Accounts Management, Authentication, Report Management, Generation, and Migration.
The website provides different accounts such as system administration (for system
management and maintenance), administrator (for provincial director and authorized access
control), and user (for local police stations). The system also provides public access to crime
maps and several other websites which do not require an account which this study focuses
on user acceptance evaluation.

Figure 4. System Architecture of Online Crime Management Information System

The system architecture as seen in Figure 4 was conceptualized and formulated by the
project team based on the identified system requirements from the initial coordination. As
seen in the figures users are categorized into 3 groups. The System Administrator (for
system management and maintenance), the Super User (for provincial director and
authorized access control) are illustrated in Figure 5, and the User (for local police stations)
is shown in Figure 6 and Figure 7. The system also provides public access to crime maps and
crime information and statistics as reflected in Figure 8 which does not require an account
in which this study focuses on user acceptance evaluation.
 The following figure demonstrates several components of the website dedicated to
different users.

Figure 5. Screenshot of the Super User Account – Crime Map

Figure 6. Screenshot of the User Account – Data Entry

 Figure 7. Screenshot of the User Account – Crime Report Statistics

Figure 8. Screenshot of the Public Access – Crime Map

Figure 8 illustrates the integration of a decision tree-generated model into web

development using React. As shown, statistics represented in percentage that the most
dominant crimes in Laguna by 24.24% are Theft and Physical Injury. The geo-map shows the
severity of Theft in the entirety of the region. From which testing and evaluation of user
acceptance are generated.
Testing and User Acceptance Evaluation

The alpha testing was conducted each scrum model was delivered it was evaluated
and tested by the researchers and programmer which resulted in a continuous room for
improvement. On the public access view of the Crime, map deliverables were tested and
deemed to be producing a precise spatiotemporal analysis based on the developed model.
However, in the System Administration part specifically, the Super User Account and User
account for the Municipality and City Police Station significant findings were found to be
improved and enhanced as the project persists. The researchers have several challenges in
challenges and issues regarding the development of this component of the website: (1) the
COVID-19 pandemic has halted the researchers from physically visiting the PNP Provincial
Office which is required in the development of the module (2) The frequent changes in the
designated of the Provincial Director since the project started in 2019. However,
improvements in the project like support from the Municipal Police Station have already
been expressed to acquire the website and assist the researchers to improve the website.

The study focuses on the user acceptability evaluation of the public access view to
determine user behavioral intention to use and acceptance of the developed website, as
well as to solicit respondents' comments and opinions regarding the website developed by
the researcher. The 45 respondents ranged in age from 19 to 35 years and were largely
students, instructors of information technology, government employees, and
entrepreneurs from Laguna's 4th district – Sta Cruz, Pila, Paete, Los Baos, Pangil, Calamba,
Nagcarlan, Mabitac, Siniloan, and Sta. Maria.

Table 4 shows the user behavior and acceptance of the developed website, as
reflected in the criteria Attitude Towards Using the Website specifically efficiency which
leads to the statement "Efficiency: The website is reliable in terms of getting the correct
spatial and temporal crime information and appropriate crime information for public
viewing." had the highest mean rated of 6.29 which can be verbally interpreted as
Acceptable however the lowest mean rated of 5.91 which falls under the Quality Factors
criteria which pertains to the statement “Reliability: The website has all the necessary
functions of the agency that responded appropriately and can be functional in normal
operations which also support on the open-ended response of the respondents “The
website is simple and user-friendly. A little suggestion in the presentation of data. I suggest
using a tabular and graphical representation of Data.”
 Table 4. User Acceptance Evaluation Result
Criteria Mean SD Verbal
Interpretation
QUALITY FACTORS
1. Reliability 5.91 0.97 Acceptable
2. Efficiency 6.02 0.99 Acceptable
3. Integrity 6.04 0.95 Acceptable
PERCEIVE EASE OF USE
1. Learnability 5.98 1.03 Acceptable
2. Understandability 6.11 0.96 Acceptable
3. User Interaction 1.10 Acceptable
6.07
Aesthetics
PERCEIVED USEFULNESS
1. Operability 6.02 0.97 Acceptable
2. Usability 6.16 1.09 Acceptable
3. Functional Completeness 6.04 1.02 Acceptable
ATTITUDE TOWARDS USING THE WEBSITE
1. Suitability 6.22 0.90 Acceptable
2. Usability 6.20 0.89 Acceptable
3. Efficiency 6.29 0.87 Acceptable
BEHAVIORAL INTENTION TO USE
1. Suitability 6.07 1.05 Acceptable
2. Attractiveness 6.09 0.95 Acceptable
3. Adaptability 6.11 0.93 Acceptable
EXPERIENCE
1. Operability 6.02 1.06 Acceptable
2. Stability 5.98 1.01 Acceptable
3. Suitability 6.13 1.04 Acceptable

CONCLUSIONS AND RECOMMENDATIONS

The project establishes a key process for generating an idea that will benefit the
community. As a result, the creation of a crime analysis website to provide policy
recommendations should be undertaken. It was discovered that predictive crime analytics
can significantly assist the PNP in developing strategic plans based on historical data. In
terms of crime forecasting, it is found that decision tree algorithms can assist in
anticipating trends of location and criminal incidents. Additionally, TensorFlow and Keras
were beneficial in developing the predictive model, while React was beneficial in
integrating the prediction model into the website. As for the crime analysis website - the
crime map for public access user review, which ultimately ended in acceptable, however,
the researchers agree with the respondents' comments and suggestions that the crime
map should include a variety of graphical representations such as a table ranking the crime
rates from highest to lowest, and a pie graph showing the comparable data of analytics
crime per towns and cities to make the system more interesting to any type of user.

Derived from the results, the following recommendations are generated:

1. It is recommended that the PNP-Laguna, municipal police stations, and the LSPU
collaborate effectively together to continue improving the crime analysis website,
primarily because the PNP-Laguna would benefit the most from the project. The
crime analysis website not only covers crime management information systems
across police departments for decentralization but also can provide analytics for
analytic purposes.

2. It is strongly recommended that Laguna develop and implement an Integrated

Crime Information Management System, which will include additional features such
as an online crime reporting system for specific localities and an intelligent
conversational application that will empower citizens in this time of pandemic the
future.

3. Finally, continued experimentation on index and non-index crime datasets to

develop intelligent systems capable of forecasting must be explored, as the index
crime dataset demonstrated a highly good result in the project.

ACKNOWLEDGEMENT

The researchers would like to extend their sincerest gratitude to the College of
Computer Studies of Laguna State Polytechnic University System for providing the
necessary support system to accomplish this research undertaking.

REFERENCES

Abdulrahman, N., & Abedalkhader, W. (2017). K-NN classifier and Naïve Bayse classifier for
crime prediction in San Francisco context. International Journal of Database
Management Systems, 9(4), 1-9.
Agrawal, S., & Agrawal, J. (2015). Survey on anomaly detection using data mining
techniques. Procedia Computer Science, 60, 708-713.
Ahishakiye, E., Omulo, E. O., Taremwa, D., & Niyonzima, I. (2017). Crime Prediction Using
Decision Tree (J48) Classification Algorithm. International Journal of Computer and
Information Technology, 6(3), 188-195.
Aldossari, B. S., Alqahtani, F. M., Alshahrani, N. S., Alhammam, M. M., Alzamanan, R. M., &
Aslam, N. (2020, January). A comparative study of decision tree and Naive Bayes
machine learning model for crime category prediction in Chicago. In Proceedings of
2020 the 6th international conference on computing and data engineering (pp. 34-38).
ACM. doi:10.1145/3379247.3379279
Amor, N. B., Benferhat, S., & Elouedi, Z. (2004, March). Naive Bayes vs decision trees in
intrusion detection systems. In Proceedings of the 2004 ACM symposium on Applied
computing (pp. 420-424).
Arya, S., Mount, D. M., Netanyahu, N. S., Silverman, R., & Wu, A. Y. (1998). An optimal
algorithm for approximate nearest neighbor searching fixed dimensions. Journal of
the ACM (JACM), 45(6), 891-923.
Baculinao, R. H., & Ceballos, R. F. (2019). An analysis of the Location and Type of Index Crimes
in the Philippines. Retrieved from https://psa.gov.ph/sites/default/files/8.6.2
Brownlee, J. (2014). Classification accuracy is not enough: More performance measures you
can use. Retrieved from https://machinelearningmastery.com/classification-accuracy-
is-not-enough-more-performance-measures-you-can-use/
Chen, H., Chung, W., Xu, J. J., Wang, G., Qin, Y., & Chau, M. (2004). Crime data mining: a
general framework and some examples. Computer, 37(4), 50–56
doi:10.1109/mc.2004.1297301
Chauhan, C., & Sehgal, S. (2017). A review: Crime analysis using data mining techniques and
algorithms. 2017 International Conference on Computing, Communication, and
Automation (ICCCA). doi:10.1109/ccaa.2017.8229823
Collier, K. (2012). Agile Analytics: A value-driven approach to business intelligence and data
warehousing. Addison-Wesley.
Chan, J. (2001). The Technology game: How information technology is transforming police
practice. Journal of Criminal Justice, 1, 139-159.
Delima, A. J. P. (2019). Applying data mining techniques in predicting index and non-index
crimes. International Journal of Machine Learning and Computing, 9(4), 533-538.
Ghoneim, S. (2019). Accuracy, recall, precision, F-score & specificity, which to optimize on?
Medium. Retrieved from https://towardsdatascience.com/accuracy-recall-precision-f-
score-specificity-which-to-optimize-on-867d3f11124
Hassani, H., Huang, X., Silva, E. S., & Ghodsi, M. (2016). A review of data mining applications
in crime. Statistical Analysis and Data Mining: The ASA Data Science Journal, 9(3), 139-
154.
Han, J., & Kamber, M. (2006). Data mining: concepts and techniques, 2nd. The University of
Illinois at Urbana Champaign: Morgan Kaufmann.
Harris, C. J. (2007). Police and Soft Technology: How Information Technology Contributes
to Police Decision Making (From The New Technology of Crime, Law and Social
Control, P 153-183, 2007, James M. Byrne and Donald J. Rebovich, eds.--See NCJ-
218026).
Ippolito, A., & Lozano, A. C. G. (2020). Tax Crime Prediction with Machine Learning: A Case
Study in the Municipality of São Paulo. In ICEIS (1) (pp. 452-459).
doi:10.5220/0009564704520459
Ivan, N., Ahishakiye, E., Omulo, E. O., & Taremwa, D. (2017). Crime Prediction Using Decision
Tree (J48) Classification Algorithm. International Journal of Computer and Information
Technology, 6(3), 188-195.
Iqbal, R., Panahy, P. H., Murad, M. A., Mustapha, A., & Khanahmadliravi, N. (2013). An
Experimental Study of Classification Algorithms for Crime Prediction. Indian Journal
of Science and Technology, 6(3), 4219-4225.
Jantan, H. & Jamil, A. Association Rule Mining Based Crime Analysis using Apriori Algorithm.
International Journal of Advanced Trends in Computer Science and Engineering, 8(1.5),
18-24. Retrieved from https://doi.org/10.30534/ijatcse/2019/0581.52019 13.
Jo, H., & Yoon, Y. I. (2018). Intelligent smart home energy efficiency model using artificial
TensorFlow engine. Human-centric Computing and Information Sciences, 8(1), 1-18.
Kargupta, H., Liu, K., & Ryan, J. (2003, June). Privacy-sensitive distributed data mining from
multi-party data. In International Conference on Intelligence and Security
Informatics (pp. 336-342). Springer, Berlin, Heidelberg.
Larman, C. (2012). Applying UML and patterns: an introduction to object-oriented analysis
and design and iterative development. Pearson Education India.
Abadin, M., Barham, P., Chen, J., Chen, Z., Davis, A., Dean, J., Devin, M.,..., & Zheng, X. (2016).
TensorFlow: A system for large-scale machine learning. In OSDI '16: Proceedings of the
12th USENIX conference on Operating Systems Design and Implementation (pp. 265-
283). ACM.
Quinlan, J. (2014). C4. 5: programs for machine learning. Elsevier.
Reichert, K. (2001). Use of information technology by law enforcement: Promising
Approaches to Addressing Crime Series. In the University of Pennsylvania, Jerry Lee
Center of Criminology, Forum on Crime and Justice.
Roth, R. E., Ross, K. S., Finch, B. G., Luo, W., & MacEachren, A. M. (2013). Spatiotemporal
crime analysis in US law enforcement agencies: Current practices and unmet
needs. Government Information Quarterly, 30(3), 226-240.
Sapin, S. B., Lerios, J. L., Padallan, J. O., Buama, C. A., & Asor, J. R. (2021). Fire incidents
visualization and pattern recognition using machine learning algorithms. Indonesian
Journal of Electrical Engineering and Computer Science, 22(3), 1427-1435.
doi:10.11591/ijeecs.v22.i3
Schwaber, K. (2004). Agile Project Management with Scrum (Developer Best Practices) 1st
Edition. Microsoft.
Schwaber K, & Beedle M. (2001). Agile Software Development with Scrum. Prentice-Hall:
Upper Saddle River.
Sukanya, M., Kalaikumaran, T., & Karthik, S. (2012). Criminals and crime hotspot detection
using data mining algorithms: clustering and classification. International Journal of
Advanced Research in Computer Engineering and Technology, 1(10), 225-227.
Thongsatapornwatana, U. (2016). A survey of data mining techniques for analyzing crime
patterns. In 2016 Second Asian Conference on Defence Technology (ACDT), 123-128.
Webb, G. I., Keogh, E., & Miikkulainen, R. (2010). Naïve Bayes. Encyclopedia of Machine
Learning, 15, 713-714.
Welsh, B. C. (2002). Technological innovations for policing: Crime prevention as the bottom
line. Cahiers Polities Studies Jaargang, 2(1), 129-132.
Zhang, H. (2004). The optimality of Naive Bayes. Retrieved from
https://www.aaai.org/Papers/FLAIRS/2004/Flairs04-097.pdf
Williamson, D., McLafferty, S., McGuire, P., Ross, T., Mollenkopf, J., Goldsmith, V., & Quinn,
S. (2001). Tools in the spatial analysis of crime. In A. Hirschfield and K. Bowers (Eds.)
Mapping and analyzing crime data (pp. 187-202). Taylor & Francis.
Wibowo, A. H., & Oesman, T. I. (2020, February). The comparative analysis on the accuracy
of k-NN, Naive Bayes, and Decision Tree Algorithms in predicting crimes and criminal
actions in Sleman Regency. In Journal of Physics: Conference Series (Vol. 1450, No. 1, p.
012076). IOP Publishing. doi:10.1088/1742-6596/1450/1/012076.