Maximizing Energy Savings in Smart Homes

through Artificial Neural Network Based Artificial

Intelligence Solutions
1 2 3 4,5,
Walaa Alayed , Aamir Akhtar , Waqar ul Hassan and Ahmed Zeeshan *

1 Department of Information Technology, College of Computer and Information Sciences, Princess Nourah bint
Abdulrahman University, P.O. Box 84428, Riyadh 11671, Saudi Arabia; wmalayed@pnu.edu.sa

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

2 Department of Avionics Engineering, College of Aeronautical Engineering, NUST, Islamabad, Pakistan;
aamirrakhtarr86@gmail.com
3 Department of Mathematics, Government College University, Lahore, Pakistan; w.hassan@gcu.edu.pk
4 Department of Mathematics and statistics, FoS, International Islamic University, Islamabad, 44000 Pakistan;

t
ahmad.zeeshan@iiu.edu.pk
5 Department of Mathematics, College of Science, Korea University, 145 Anam-ro, Seongbuk-gu, Seoul 02841,

ip
Republic of Korea

* Correspondence: Ahmed.zeeshan@iiu.edu.pk ,

cr
Funding: This research was funded by Princess Nourah bint Abdulrahman University

us
Researchers Supporting Project number (PNURSP2024R 500), Princess Nourah bint
Abdulrahman University, Riyadh, Saudi Arabia.
an
Institutional Review Board Statement: Not applicable

Informed Consent Statement: Not applicable

M

Data Availability Statement: The data that support the findings of this study are openly
available at https://traces.cs.umass.edu/index.php/smart/smart.
d

Acknowledgments: The authors would like to acknowledge Princess Nourah bint

e

Abdulrahman University Researchers Supporting Project number (PNURSP2024R500),

pt

Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia for supporting this
project.
ce

Conflicts of Interest: The authors declare no conflicts of interest.

Ac

© The Author(s) 2025. Published by Oxford University Press on behalf of National Institute of Clean-
and-Low-Carbon Energy
This is an Open Access article distributed under the terms of the Creative Commons Attribution-
NonCommercial License (https://creativecommons.org/licenses/by-nc/4.0/), which permits non-
commercial re-use, distribution, and reproduction in any medium, provided the original work is
properly cited. For commercial re-use, please contact reprints@oup.com for reprints and translation
rights for reprints. All other permissions can be obtained through our RightsLink service via the
Permissions link on the article page on our site—for further information please contact
journals.permissions@oup.com.
Abstract: In designing a modern home with focus on comfort of resident and energy usage
optimization simultaneously, the rise of the Internet of Things and incorporation with sensors
technology plays a vital role these days. The first and foremost task is to predict the energy
consumption in based on available data. This study investigates the integration of artificial
neural networks in smart home technology to improve energy usage prediction and
efficiency, with compromising the comfort of occupant. A dynamic model based on an

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

artificial neural network model is designed in this study which artificially controls light,
heating process and cooling to cut down energy wastage. Data from 114 single-family
apartments for energy consumption is collected from year 2014 to 2016. Energy consumption

t
ip
is predicted by current model with an accuracy of up to 99.9% for energy usage patterns.
Which helps to optimizing resource management in real time. A robust modeling approach

cr
i.e. multi-layer perceptron networks was implemented along with energy usage data. 70% of
data is used for training the neural networks and rest for testing and validation purpose. The

us
current defined model shows a significant improvement in prediction accuracy of energy
usage and efficiency when compared to state-of-the-art models. Metrics such as R-values and
an
mean square error are employed to check accuracy. These results show the essential role of
artificial intelligence in improving energy management for smart buildings, with potential
M

benefits including significant energy usage and loss management to help improve sustainable
living.
d

Keywords: Artificial Neural Networks; Artificial intelligence (AI); Energy management;

e

Internet of things (IoT); smart home technology; Application of AI

pt
ce
Ac
 Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025
t
ip
cr
us
an
Graphical abstract

M
d e
pt
ce
Ac
1. Introduction

A smart home is a place where all the appliances are networked together such that a
centralized, but controlled, and monitored system is created to run this place [1]. The Internet
of Things (IoT) enables this arrangement by joining devices to one common link and support
data sharing [2] which helps to know the choose of resident. Since, smart home technology,

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

incorporate different components of smart home into one central unit, which can be
monitored and controlled according to requirement of resident, becomes progressively
important. If early smart home technology is evaluated it becomes imperative that focus of

t
ip
smart homes was mainly on automation of appliances and ease of the resident. But this trend
quickly shaped up and energy conservation and cost reduction become more important

cr
components. With the progression of processing power, smart homes have evolved from

us
mere connected networks to AI-driven systems that provide personalized services based on
individual‘s habits [3].
This study addresses the critical issue of optimizing smart home operations using artificial
an
neural networks (ANN). Artificial intelligence (AI)-powered smart homes can adapt and
learn from data provided, enhance its efficiency and optimize user experience when
M

compared to old technologies. Many household appliances are AI-powered such as security
cameras, fans, lights, heating, ventilation and cooling (HVAC) systems, and water heating
d

appliances etc [4]. Despite significant improvement in AI applications globally, in fields like
e

data mining, computer vision, and natural language processing, there remains a considerable
pt

gap in usage of AI to optimize smart home environments comprehensively.

ce

Research on artificial intelligence started right after World War II. Turing first took the
initiative in AI research in 1947 and published a paper in 1950 [5] discussing the possibility
Ac

of machine intelligence. Consequently, many new scholars entered this subject [6-11].
Almost every field including data mining, computer vision, speech recognition, video games,
natural language understanding, and expert system building has been engulfed by artificial
intelligence.
In the realm of artificial intelligence, a ANN is a technique that teaches machines to interpret
data just like a human brain [12]. The objective of such neural networks is to predict and
solve complex issues with high accuracy without requiring human interference [13]. Neurons
are the basic building blocks of an ANN-based intelligent system. In a ANN there are three
basic layers, an input layer, one or more hidden layers, and an output layer [14]. Input layer
takes data from the outside world to learn which further goes to hidden layers for processing
and then the final output is sent back to the output layer [15] as shown in Figure 1. Any
output from the output layer is the artificial neural networks' reaction to the supplied input
data.

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

t
ip
cr
us
an
Figure 1. A simple ANN architecture [16].
M

ANN nodes accept input signals, process them using activation functions by the hidden layer
d

nodes, and then compute the final output by processing the results of the hidden layer [16] as
shown in figure 2.
e
pt
ce
Ac

Figure 2. Input output structure of neuron in ANN system [16].

Our research aims to bridge this gap by suggesting a novel ANN-based model to predict and
solve energy efficiency issues in smart homes with high precision and minimal human
interference. The key contributions of this work are
 To build an efficient model based on ANN for smart home energy management.
 To demonstrating the superiority of purposed model over previous state-of-the-art
(SOTA) methods via testing and validation.
 To enhance the accuracy and optimal operational performance of smart home systems by
minimizing energy usage and enhancing comfort of resident.

1.1. Literature Overview

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

Several studies have investigated AI-driven approaches to energy management in smart
homes, focusing on energy efficiency and occupant comfort.

t
Kim [17] proposes an Internet of Things (IoT)-based home energy management system

ip
(HEMS) which emphasizes on energy efficiency and occupant satisfaction. The analysis

cr
focus on the need and importance of energy management system in residential buildings. The
authors of the paper used three estimation methods; derivation of comfort temperature,

us
device-free sleep prediction, and occupancy-probability-based outing prediction, and
proposed four heating control strategies based on this process: outing, occupancy, comfort,
an
and sleep-based control. By utilize IoT, the proposed HEMS model tends to reduce Energy
consumption via real time control of home appliances. The paper effectively shows the
potential of IoT in improving energy management in residential home. However, the model's
M

scalability and real-time adaptability to fluctuating energy demands were not addressed,
limiting its application in broader smart home environments.
d

Lissa et al. [18] investigate heat pump control using deep reinforcement learning (DRL) to
e

improve microgrid energy efficiency by optimizing photovoltaic energy use. The study
pt

proposes an algorithm for new dynamic indoor temperature set-point adjustment which
provides more flexibility and savings on photovoltaic (PV) self-consumption. In this study
ce

using transfer learning the convergence is enhanced while reducing training time
significantly. The numerical experiments identifies that the proposed DRL algorithm
Ac

combined with dynamic content saves energy by 8% on average and up to 16% in summer.
The results highlight the potential DRL-based control systems in improving energy efficiency
in micro grids. Since the algorithm is calibrated not to exceed 1% of the set temperature, user
comfort is not affected. In terms of demand-side management, DRL management exhibits
excellent performance with more than 10% operating flexibility by predicting and delaying
the self-optimization of PV use. Although its reliance on photovoltaic energy may restrict its
applicability in homes without solar infrastructure.
Ahmed et al. [19] presents a deep learning approach that provides a unique distribution
mechanism for wireless sensor IoT networks while improving energy efficiency and data
optimization. In this case, EE (energy efficiency) and SE (spectrum efficiency) are two
competing optimization objectives. The energy efficiency of the network is improved thanks
to deep neural networks based on whale optimization. A heuristic-based multi-objective
firefly algorithm is used to optimize the data. The optimal resource allocation is achieved by

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

minimizing the overall transmission and the optimal relay selection is achieved by the quality
of service (QoS) criteria. Simulation results show that this model has 96% matching rate,
95% power efficiency, 75% quality of service, 85% spectral efficiency and 91% network

t
ip
lifetime rate for the traditional models.
A study by Alzoubi [20], explores data fusion technology to enhance energy optimization in

cr
building management systems, achieving a high estimation accuracy of 92%.the energy
optimization problem is tackled again by the energy management method. Recently, there has

us
been interest in data fusion in the context of building energy efficiency. The accuracy and
error of energy estimation are calculated using the data fusion technology which is proposed
an
in this study. The simulation results are compared with previous studies and reached an 92%
estimation accuracy which is a very high result. Unlike other studies that focus on the use of
M

AI technology in buildings or individual apartments with local offices and data storage,
Alzoubi [20] have worked to control heat in three types of buildings and analyzed the
d

effectiveness of AI technology. Energy savings of approximately 20% to 40%, measures for

e

frequent ventilation of the buildings have been made, even in cases with older-type heating
systems. However, despite its accuracy, the method's effectiveness in real-time applications
pt

or in environments with varying data quality remains uncertain, indicating a potential

ce

limitation in dynamic smart home systems. In contrast to previous research, In this article,
we present an ANN-based IoT smart home solution with much higher efficiency than all
Ac

previous work done in the same field. Our model aims to use an ANN model to that achieves
higher efficiency by dynamically optimizing energy use for lighting, heating, cooling,
ventilation, and plumbing. Using open-source data, our ANN model achieved 99.99%
accuracy in predicting energy consumption patterns, offering a robust solution that addresses
real-time, adaptable energy management for smart homes. This high accuracy suggests
significant potential for energy savings and improved sustainability in residential settings,
optimize energy consumption in smart homes through various means by controlling time to
switch lights on or off, control heating and ventilation for optimum use of energy. Our paper
uses an open-source data to train an AI model and presents its accuracy as verification of
model accuracy.
The paper by Syamala et al. [21] explores combination of machine learning technique with
IoT for smart home energy efficiency management. The authors discuss the importance of
accuracy in the prediction of energy for urban planning. This study employs deep learning
techniques for energy efficiency in smart homes, identifying the optimal window size in order

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

to predict accurately.
Hussain et al. [22] propose improvement to HEMS by defining a multi-stage energy
optimization and trading process in smart homes. The framework also includes mixed integer

t
ip
linear programming (MILP) to schedule energy efficiently. The proposed techniques improve
energy saving 38.85% in peak times and also reduce operational costs.

cr
Umair et al. [23] present a probabilistic model based on a Markov chain, to predict energy
consumption patterns defining algorithm called PF-PEC1. Also, the study highlights fog-base

us
computing and its application. It shows up to 36% energy conservation.
Many recent works [24-28] and work mentioned therein are important innovation in this
an
field. The studies show many models which integrated AI techniques with smart home
environment defining different purpose and optimizing energy. The energy optimality and
M

prediction of usage when due to uninvited change is very important AI need react and decide
how to optimize the energy. To build a sustainable residence when talking of energy needs
d

accuracy of prediction is very essential. Kim [17] and Alzoubi [20] show the effectiveness of
e

IoT to enhance energy efficiency. Whereas, Ahmed et al. [19] and Hussain et al. [22] show
usage of AI techniques in improved energy management.
pt
ce

However, a gap persists in in predicting the availability of a comprehensive method that not
only identifies energy consumption patterns efficiently but also employs AI optimization to
Ac

achieve substantial reductions in both energy usage and electricity costs. While individual
studies have made strides in specific areas—such as accurate energy prediction, user comfort,
and distribution network constraints—there remains an unmet need for an integrated
approach that holistically addresses these challenges. Developing such a method would
bridge the current gaps, providing a robust solution for smart home energy management that
maximizes efficiency and minimizes costs.
This overview underscores the importance of further research and innovation in creating an
all-encompassing AI-driven framework that leverages the strengths of various methodologies
to optimize energy management comprehensively.

2. Research Methodology

2.1. Problem Definition

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

The rising complexity in smart home technology renders identification of new challenges.
First and foremost is energy management of interconnected IoT appliances while upholding

t
resident‘s comfort. Although IoT assists in seamlessly collecting data and transmitting it

ip
across all devices, it lacks the ability to optimize energy consumption, fuel management and

cr
efficient interpretation of data in real time across the entire network. A critical gap in the
literature is the lack of a robust and integrated AI system which is capable of processing

us
extensive real-time data and making decisions to minimize energy usage without
compromising comfort of user.
an
Keeping this in view, an advanced AI-based solution is required which monitors, manages,
and decide based on the data of usage patterns optimizes the energy consumption. The
M

current study focuses on purposing an AI-based approach using ANNs of energy distribution
for smart home technology and aims to significantly reduce wastage of energy without
d

compromising user comfort.

e

2.2. Working Principle

pt

2.2.1. Data Collection and Preprocessing

ce

The initial step in addressing the problem involves gathering comprehensive datasets. We
utilized open-source electricity data that recorded the electricity consumption of 114 single
Ac

households between 2014 and 2016 [29]. In table 1, time represents the number of seconds
since 2016-01-01 00:00:00, and power consumption represents power consumed in kilowatts
during this time. Figure 3 shows how energy consumption has changed over time,
highlighting differences in energy consumption patterns. The main goal of this stage is to
understand the characteristics of the data and identify potential patterns or inconsistencies.
To enhance the dataset for model training, in table 2, we introduced a third column, which
categorized energy consumption into three labels: low, medium, and high. Specifically,
power usage between 0 and 0.8 kW was labeled as ‗0‘ (low consumption), between 0.8 and 3
kW as ‗1‘ (medium consumption), and above 3 kW as ‗2‘ (high consumption). This
transformation allowed the data to be structured for classification tasks in the machine
learning model.

Table 1. Dataset sample

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

Power Consumption (kW) Timestamp (Seconds)
0.682950 2016-01-01 00:01:00

t
0.682983 2016-01-01 00:02:00

ip
0.682383 2016-01-01 00:03:00

cr
0.682283 2016-01-01 00:04:00

us
an
M
e d
pt

Figure 3. Energy Consumption VS. Time-stamps.

ce

Table 2. Dataset sample with labels

Power Consumption (kW) Timestamp (Seconds) Label

Ac

0.682950 2016-01-01 00:01:00 0

0.682983 2016-01-01 00:02:00 0
0.682383 2016-01-01 00:03:00 0
0.682283 2016-01-01 00:04:00 0
2.2.2. Data Cleaning and Splitting
The next step is data cleaning, which involves removing all NaN values. Interestingly, since
the data set is already clean, no data is deleted during this process. The data is then split 80:20
into training and validation, resulting in 403,019 training data and 100,755 valid data. This
separation is necessary to ensure that models can be trained on large and efficient datasets.

2.2.3. Designing the ANN Model

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

After cleaning the data, the next step was to design our ANN model which posed significant
challenges regarding the number of layers and neurons required for optimal accuracy. The

t
process commenced by cleaning the data for any non-number or empty cell present in the

ip
tabulated data. It was then split into training data and test data to train and validate our model
on the same data respectively. Sequential models were employed to build high-level features

cr
through successive layers [30].

us
Our model consists of seven layers, beginning with a dense layer containing 2056 neurons. In
a neural network, a dense layer is also referred to as a fully connected layer serves as a
an
fundamental building block of any ANN-based system. It is termed "dense" because each
neuron in the layer is connected to every neuron in both the preceding and following layers.
The output of a dense layer neuron is computed by taking the weighted sum of its inputs and
M

adding a bias term. This weighted sum is then processed through an activation function to
introduce non-linearity, which enables the network to model complex relationships within the
d

data (see figure 4).

e

Mathematically, the output of the -th neuron in the dense layer can be expressed as [31].
pt

(1) (∑ )
ce
Ac

where:

1. represents the input to the neuron,

2. represents the weight associated with input xj,
3. is the bias term,
4. is the activation function applied to the weighted sum.
 .
.
.
.
Input layer

Dense layer 1

Batch Normalization

Dropout

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

Dense layer 2

t
ip
Dense layer 3

cr
Batch Normalization

us
Dropout

Output layer
an
Figure 4. Proposed ANN model architecture.
M

2.2.2.1. Activation Function: Rectified Linear Unit

In this model, the rectified linear unit (ReLU) was used as the activation function in the first
d

dense layer. ReLU is the most widely used activation function in deep networks due to its
e

simplicity and effectiveness. It introduces non-linearity by outputting the input directly [32] if
pt

it is positive and zero otherwise. This allows the model to learn complex patterns and
relationships in the data. ReLU is mathematically defined as [33].
ce

(2) ( ) ( )
Ac

2.2.2.2. Layer Configuration

The first dense layer of the model uses ReLU as activation function with 2056 neurons. The
second layer also, employees ReLU as the activation function, but neuron count is 1024
neurons. To improve training stability and speed up convergence, a third layer is introduced
for batch normalization. In the training process this layer helps to normalize the outputs from
the previous layer at each batch, hence in the process it reduces covariant shift to ultimately
speed up learning. The fourth layer, known as the dropout layer, helps to prevent overfitting
of data. Here, some of the input units are set as zero randomly, which enforces the model to
learn redundant representations. This will result in enhancing the ability of the model to
handle the unseen data by stopping over dependence on particular neurons.
The fifth and sixth layers are again dense layers comprise 512 and 256 neurons, respectively.
Also, both layers use the ReLU as activation functions. These help to learn abstract
topographies from the imparted data.
The finally, the output layer of the model has three neurons which signifies low, medium and

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

high power consumption.
2.2.2.3. Output Layer and SoftMax Activation

t
The output layer uses the SoftMax activation function. SoftMax function is designed so that it

ip
performs tasks classified as multi-class. It transforms the output of the model into a
probability distribution and assign a probability value to all three out pots i.e. high medium

cr
and low.

us
Mathematically, given an input vector ( ) the SoftMax function computes
the output vector ( ) [34], where:
an
(3) ∑
,
M

Sum of these probabilities become 1.

These steps ensures that the model inputs are categorized as the required output with high
d

precision and accuracy across all kind of known and unknown data.
e

3. Model Training and Validation

pt

In this study, the model was trained and validated for the energy consumption in an apartment
ce

with single accommodation and displayed encouraging results. This process generates a
training accuracy of up to 99.96% and a validation accuracy of almost 99.99%, which shows
Ac

that the current model is extremely reliable and accurate.

The model was trained using Google Colab's T4 GPU, [35] with each epoch taking
approximately 645 seconds and involving 12,595 iterations. Over 15 epochs, the accuracy
continuously improved while training and validation losses decreased, as shown in Table 3.
From table 3, it is evident that the model's accuracy improved with each epoch, while the
training and validation losses consistently decreased.
 Table 3. Model training summary

Validation Loss Validation Accuracy Training Loss Training Accuracy Epoch

0.0134 0.9982 0.0409 0.9864 1
0.0144 0.9934 0.0138 0.9954 2
0.0069 0.9965 0.011 0.9964 3
0.0057 0.998 0.0098 0.9967 4

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

0.0195 0.9954 0.0089 0.9969 5
0.0062 0.9966 0.0082 0.9971 6

t
0.008 0.9971 0.0079 0.9974 7

ip
0.0015 0.9998 0.0033 0.9988 8

cr
0.0011 0.9998 0.0021 0.9993 9
9.57E-04 0.9996 0.0016 0.9994 10

us
8.83E-04 0.9997 0.0015 0.9994 11
7.51E-04 0.9998 0.0011 0.9995 12
an
5.76E-04 0.9999 0.0012 0.9996 13
5.54E-04 0.9999 0.0012 0.9996 14
5.09E-04 0.9998 0.0011 0.9996 15
M

4. Results
d

To visualize the training process, several charts were plotted. Figure 5 compares training vs.
e

validation accuracy, and Figure 6 compares training vs. validation loss. These charts
pt

highlight the model's learning pattern, showing fluctuations in the early epochs, followed by
steady improvements in both accuracy and loss. There is a noticeable rise in accuracy and a
ce

gradual decline in loss, demonstrating the effectiveness of the model's architecture and
training process.
Ac
 Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025
t
ip
Figure 5. Training vs Validation Accuracy

cr
us
an
M
e d
pt

Figure 6. Training vs Validation Loss

ce

Further analysis was performed using histograms for both accuracy and loss. As seen in
Figure 7 (accuracy histogram) and Figure 8 (loss histogram), the accuracy starts lower but
Ac

progressively improves, while the loss shows a consistent decrease. This indicates that the
model is converging well, and overfitting is avoided due to techniques like dropout layers.
 Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025
t
ip
cr
Figure 7. Accuracy histogram

us
an
M
e d
pt

Figure 8. Loss histogram

ce

5. Comparison with SOTA Methods

Ac

To validate the effectiveness of our ANN-based model, we performed rigorous testing and
comparison with existing SOTA methods. Training and validation were performed for energy
consumption data taken from a single apartment, providing very impressive results. The
model achieved a max training accuracy of 99.96% and a validation accuracy of 99.99%,
showing high reliability and precision. We consider following SOTA methods for
Comparison purpose

 AI-based HEMS [17]: Focused on occupant satisfaction and energy efficiency,

achieving 14% energy savings and 91% satisfaction. Our model's higher prediction
 accuracy could enhance such systems by providing precise forecasts, thereby improving
energy management decisions.

 DRL Algorithm for Energy Management [18]: Achieved up to 16% energy savings in
summer and 9.5% improvement in energy efficiency. Our model's superior accuracy
provides a reliable basis for forecasting future energy consumption, further optimizing
the DRL approach.

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

 Deep Learning for IoT Networks [19]: Achieved a 96% matching rate and 95% power
efficiency. Our model focus on energy consumption will help to improve the [19] results

t
ip
by incorporating ANN to optimize energy as discussed in current study.

cr
 Alzoubi's Study [20]: uses data fusion-based model with energy management method he
achieve an accuracy of 92%. Proposed model is shows much better accuracy with

us
training and validation i.e. 99.96% and 99.99%, respectively.
an
 Umair et al. [21]: uses a fog-based IoT system for smart home energy efficiency
management, they demonstrated up to 36% energy savings. In comparison, our model
demonstrated more accurate predication both with training and validation and hence
M

more reliable prediction model for energy consumption patterns.

 Hussain et al. [22]: developed a multi-stage optimization method for managing energy
d

and predicted 38.85% reduction in consumption in peak power. Whereas, current

e

technique, allows the continuous improvement in training and validation accuracy, due to
pt

significant decrease in loss.

 Syamala et al. [23]: integrated machine learning with IoT for accurately predicting
ce

smart home energy consumption. In the current study, we modify their method by the use
of ANN to achieve even better results and avoid overfitting.
Ac

The R-value close to 1 and mean square error (MSE) is minimal. Current ANN-based model
shows very good results, therefore, increasing the predictive accuracy and ensure efficient
working of smart home technology.
5. Conclusion
Current analysis presents an innovative way to improve consumption prediction of energy in
smart home technology using ANNs. this helps in predicting and constructing society with
more predictability and smart homes more in line with random or alien entry in the
environment. For doing so we tested and validated the current model and compared the result
shown in literature. These results showed that current ANN-based model is an improvement

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

to existing cutting-edge techniques or SOTA methods in assessing energy utility.
 The accuracy of training data is 99.96% and validation accuracy is 99.99%, which
clearly defines the reliability of current study.

t
ip
 Google Colab's T4 GPU proficiently handled enormous datasets, guaranteeing high
precision and low error rates. Graphical representation, including training versus

cr
validation accuracy, confirmed the model‘s effectiveness in learning and overfitting of

us
data.
 Comparative analysis with existing SOTA methods exhibited the current model's
prevalence.
an
 The integration of ANN into smart home technology guarantees huge progressions in
energy management and gives accurate prediction which improve processes of decision
M

by the provided technology.

e d
pt
ce
Ac
Future work: This research contributes to the evolving field of smart homes by
demonstrating how AI can revolutionize home automation and energy efficiency. Future
work may expand this study by integrating additional smart home components and exploring
real-time applications, further enhancing the capabilities and benefits of AI-driven smart
home systems.

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

Funding: ―This research was funded by Princess Nourah bint Abdulrahman University
Researchers Supporting Project number (PNURSP2024R500), Princess Nourah bint
Abdulrahman University, Riyadh, Saudi Arabia.‖

t
ip
Data Availability Statement: The data that support the findings of this study are openly

cr
available at https://traces.cs.umass.edu/index.php/smart/smart.

Acknowledgments: ―The authors would like to acknowledge Princess Nourah bint

us
Abdulrahman University Researchers Supporting Project number (PNURSP2024R500),
Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia for supporting this
an
project.‖

Conflicts of Interest: The authors declare no conflicts of interest.

M
e d
pt
ce
Ac
References

1. Banerjee, A., Sufyanf, F., Nayel, M. S., & Sagar, S. Centralized framework for controlling
heterogeneous appliances in a smart home environment. In: 2018 International Conference on
Information and Computer Technologies (ICICT), DeKalb, IL, USA, 23-25 March 2018, 78-82.
New York: IEEE. DOI: 10.1109/INFOCT.2018.8356844

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

2. Madakam, S., Ramaswamy, R., & Tripathi, S. (2015). Internet of Things (IoT): A literature
review. Journal of Computer and Communications, 3(5), 164-173.

t
DOI: 10.4236/jcc.2015.35021

ip
3. Bugeja J, Jacobsson A, Davidsson P. Smart connected homes. In: Hassan Q (ed.) Internet of
things A to Z: Technologies and applications. Hoboken, NJ, USA: John Wiley & Sons, Inc.,

cr
2018, 359-84. DOI: 10.1002/9781119456735.ch13
4. Saxena, A., Bhatt, A. K., & Kumar, R. (Eds.). (2022). Artificial Intelligence for Societal

us
Development and Global Well-Being. Hershey, PA: IGI Global Scientific Publishing.
https://doi.org/10.4018/978-1-6684-2443-8
an
5. Turing, A. (2003). Machinery and intelligence. Mind: A Quarterly Review of Psychology and
Philosophy, 59(3), 50-64.
6. van Assen, M., Muscogiuri, E., Tessarin, G., & De Cecco, C. N. (2022). Artificial intelligence: A
M

century-old story. In: De Cecco, C.N., van Assen, M., Leiner, T. (eds.) Artificial Intelligence in
Cardiothoracic Imaging. Cham: Springer International Publishing, 3-13. DOI: 10.1007/978-3-
d

030-92087-6_1
e

7. Masud, U., Ali, M., Qamar, F., Zeeshan, A., & Ikram, M. (2020). Dual mode spectroscopic
pt

biomedical sensor: Technical considerations for the wireless testbed. Physica Scripta, 95(10),
105206. DOI 10.1088/1402-4896/abb49c
ce

8. Rathor, S. K., & Saxena, D. (2020). Energy management system for smart grid: An overview and
key issues. International Journal of Energy Research, 44(6), 4067-4109. DOI: 10.1002/er.4883
9. Turing A. Computing Machinery and Intelligence. Mind: A Quarterly Review of Psychology and
Ac

Philosophy 1950; 59:433-60. DOI:10.1093/mind/LIX.236.433

10. Masud, U., Saeed, T., Malaikah, H. M., Islam, F. U., & Abbas, G. (2022). Smart assistive system
for visually impaired people obstruction avoidance through object detection and
classification. IEEE access, 10, 13428-13441. DOI: 10.1109/ACCESS.2022.3146320
11. Ahmed, B., Akram, T., Naqvi, S. R., Alsuhaibani, A., Altherwy, Y. N., & Masud, U. (2024). A
Novel Deep Learning Framework with Meta-Heuristic Feature Selection for Enhanced Remote
Sensing Image Classification. IEEE Access. 12, 91974-91998. DOI::
10.1109/ACCESS.2024.3422368
12. Dastres, R., & Soori, M. (2021). Artificial neural network systems. International Journal of
Imaging and Robotics (IJIR), 21(2), 13-25. https://hal.science/hal-03349542/
13. Zeng, J., Li, C., & Zhang, L. J. (2018). A face recognition system based on cloud computing and
AI edge for IOT. In: Edge Computing–EDGE 2018: Second International Conference, Held as
Part of the Services Conference Federation, SCF 2018, Seattle, WA, USA, June 25-30, 2018,
Proceedings 2 (pp. 91-98). Springer International Publishing. DOI: 10.1007/978-3-319-94340-4_7
14. Zeeshan, A. Khalid, N. Ellahi, R. Khan, M. I., Alamri, S. Z., (2024), Analysis of nonlinear

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

complex heat transfer MHD flow of Jeffrey nanofluid over an exponentially stretching sheet via
three phase artificial intelligence and Machine Learning techniques, Chaos, Solitons & Fractals,

t
189 (1), 115600. https://doi.org/10.1016/j.chaos.2024.115600.

ip
15. Houzé, É. (2019). Explainable Artificial Intelligence for the Smart Home: Enabling Relevant
Dialogue between Users and Autonomous Systems. (Dissertation: Master of Science in

cr
Engineering - Computer Science and Technology, School of Electrical Engineering and Computer
Science (EECS), KTH, Sweden) Retrieved from

us
https://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-252705
16. Zupan, J. (1994). Introduction to artificial neural network (ANN) methods: what they are and how
an
to use them. Acta Chimica Slovenica, 41(3), 327.
17. Kim, J. (2016). HEMS (home energy management system) base on the IoT smart
home. Contemporary Engineering Sciences, 9(1), 21-28. DOI: 10.12988/ces.2016.512316
M

18. Lissa, P., Schukat, M., Keane, M., & Barrett, E. (2021). Transfer learning applied to DRL-Based
heat pump control to leverage microgrid energy efficiency. Smart Energy, 3, 100044. DOI:
d

10.1016/j.segy.2021.100044
e

19. Ahmed, Q. W., Garg, S., Rai, A., Ramachandran, M., Jhanjhi, N. Z., Masud, M., & Baz, M.
pt

(2022). Ai-based resource allocation techniques in wireless sensor internet of things networks in
energy efficiency with data optimization. Electronics, 11(13), 2071. DOI:
ce

10.3390/electronics11132071
20. Alzoubi, A. Machine learning for intelligent energy consumption in smart homes. International
Journal of Computations, Information and Manufacturing (IJCIM) 2022, 2 (1). DOI:
Ac

10.54489/ijcim.v2i1.75
21. Syamala, M., Komala, C. R., Pramila, P. V., Dash, S., Meenakshi, S., & Boopathi, S. (2023).
Machine learning-integrated IoT-based smart home energy management system. In: Swarnalatha
P., Prabu S. (eds.) Handbook of Research on Deep Learning Techniques for Cloud-Based
Industrial IoT. Hershey, PA: IGI Global Scientific Publishing, 219-35
22. Hussain, S., Azim, M. I., Lai, C., & Eicker, U. (2023). Multi-stage optimization for energy
management and trading for smart homes considering operational constraints of a distribution
network. Energy and Buildings, 301, 113722. https://doi.org/10.1016/j.enbuild.2023.113722
23. Umair, M., Cheema, M. A., Afzal, B., & Shah, G. (2023). Energy management of smart homes
over fog-based IoT architecture. Sustainable Computing: Informatics and Systems, 39, 100898.
https://doi.org/10.1016/j.suscom.2023.100898
24. Ameur, A., Berrada, A., & Emrani, A. (2023). Intelligent energy management system for smart
home with grid-connected hybrid photovoltaic/gravity energy storage system. Journal of Energy
Storage, 72, 108525. https://doi.org/10.1016/j.est.2023.108525
25. Rigo-Mariani, R., & Ahmed, A. (2023). Smart home energy management with mitigation of

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

power profile uncertainties and model errors. Energy and Buildings, 294, 113223.
https://doi.org/10.1016/j.enbuild.2023.113223

t
26. Wei, G., Chi, M., Liu, Z. W., Ge, M., Li, C., & Liu, X. (2023). Deep reinforcement learning for

ip
real-time energy management in smart home. IEEE Systems Journal, 17(2), 2489-2499.
https://doi.org/10.1109/JSYST.2023.3247592

cr
27. Boynuegri, A. R., & Tekgun, B. (2023). Real-time energy management in an off-grid smart home:
Flexible demand side control with electric vehicle and green hydrogen production. International
Journal of Hydrogen
https://doi.org/10.1016/j.ijhydene.2023.01.239
us
Energy, 48(60), 23146-23155.
an
28. Gherairi, S. (2023). Design and implementation of an intelligent energy management system for
smart home utilizing a multi-agent system. Ain Shams Engineering Journal, 14(3), 101897.
https://doi.org/10.1016/j.asej.2022.101897
M

29. Barker, S.; Mishra, A.; Irwin, D.; Cecchet, E.; Shenoy, P.; Albrecht, J. Smart: (2012), An open
data set and tools for enabling research in sustainable homes. SustKDD: Workshop on Data
d

Mining Applications in Sustainability, Beijing, China, 12/08/2012, 111(112), 108.

e

30. Denoyer, L.; Gallinari, P. Deep sequential neural network. 2014. arXiv preprint arXiv:1410.0510
pt

2014.
31. Leo, C. The Math Behind Neural Networks. Medium. [Online] Available at:
ce

https://towardsdatascience.com/the-math-behind-neural-networks-a34a51b93873 (Accessed: 12
February 2024).
32. Hussain, S. M., Ijaz, N., Dhahbi, S., Saleem, N., & Zeeshan, A. (2024). A comparative study of
Ac

exact and neural network models for wave‐ induced multiphase flow in nonuniform geometries:
Application of Levenberg–Marquardt neural networks. ZAMM‐ Journal of Applied
Mathematics and Mechanics/Zeitschrift für Angewandte Mathematik und Mechanik,
e202400210. DOI: 10.1002/zamm.202400210
33. Brown, J. A Gentle Introduction to the Rectified Linear Unit (ReLU), Deep Learning
Performance (2020) Available at: https://machinelearningmastery.com/rectified-linear-
activation-function-for-deep-learning-neural-networks/ (Accessed: 12 February 2024)
34. C, B.P. Softmax activation function: Everything you need to know. Pinecone. Available at:
https://www.pinecone.io/learn/softmax-activation/ (Accessed: 12 February 2024).
35. Google. (2024). Google Colaboratory. Available at . https://colab.research.google.com/github/d2l-
ai/d2l-tvm
colab/blob/master/chapter_gpu_schedules/arch.ipynb#scrollTo=CVpcbZkl3NSh (Accessed: 18
February 18, 2024)

Downloaded from https://academic.oup.com/ce/advance-article/doi/10.1093/ce/zkae113/7950417 by guest on 15 January 2025

t
ip
cr
us
an
M
e d
pt
ce
Ac