Stock Market Analysis &Predictions
Preety Sharma Noopur Ridhima Gulati Jagriti
Exxxx 22BCS14084 22BCS14507 22BCS14515
University Institute of University Institute of University Institute of University Institute of
Engineering Engineering Engineering Engineering
Chandigarh University chandigarh University chandigarh University chandigarh University
Punjab, India Punjab, India Punjab, India Punjab, India
Exxxx@cuchd.in 22BCS14084@cuchd.in 22BCS14507@cuchd.in 22BCS14515@cuchd.in
Yash Riya
22bcs12563 22BCS13811
University Institute of University Institute of
Engineering Engineering
Chandigarh University Chandigarh University
Punjab, India Punjab, India
22BCS12563@cuchd.in 22BCS13811@cuchd.in
Abstract—Due to the non-linear and volatile nature of financial K-Neighbors Regressor is actually better on data concerning
markets stock price forecasting is a complex process. This is the stock market since these are generally patterns which are
because with the help of machine learning, predictive models non-linear and harder to catch in an equation like one does
have been in a position to analyze and make a good prediction
traditionally. Thus, having looked at what's been occurring
of stock prices. The purpose of this research is to design and
within a past timeline and developing a trend of behavior,
compare two regression models namely K-Neighbors Regressor
and Passive Aggressive Regressor in order to predict the closing often such models expose previously unseen trends and
price of stocks. The dataset is the historical stock data including relationships in that stock.
Open, High, Low, and Close prices to train the models. Both However, the Passive Aggressive Regressor is more of a linear
models are evaluated based on the standard error metrics such
model, which focuses more on speed and efficiency when
as Root Mean Squared Error (RMSE) and Mean Absolute
Percentage Error (MAPE). From the experimental results, it
dealing with huge datasets. That is why it becomes unique: it
can be seen that both techniques have their own advantages: K- easily learns new data, and in the world of the stock market,
Neighbors Regressor is sensitive to local patterns, whereas where the dynamics are constantly changing, that is what
Passive Aggressive Regressor is suitable for streaming data and really matters. It is very resistant to noise and outliers; this is
sudden price changes. The results of this study can help in a frequent challenge when working on real-world financial
understanding which model is more suitable for real-time data, so it makes it a perfect tool in the fast-paced world of
forecasting in the stock market. trading.
1. Introduction This research aims to evaluate the performance of these
models in predicting stock prices and underlying market
The stock market is a storm of uncertainty, given factors such
trends. We shall apply both techniques on real stock data and
as economic reports, market sentiment, even global
hopefully provide valuable insights on how machine learning
happenings. It's pretty much a rough call in predicting the
can improve decision-making and the accuracy of forecasts in
prices of stocks in the present. It challenges investors and
the finance sector. It is not only a matter of comparison of the
analysts to make the right decisions. Recently, a new
models but also their actual feasibility of being used in real-
promising tool with its rising promotion would help unveil the
time market conditions, where quick and correct predictions
complexity that is to be understood as well as ways of
are very important.
interpreting new data have taken place, making forecasts
much more accurate. This is one area of a very rapidly growing field of finance:
machine learning. The methods presented here are all
I will use for analysis of stocks trends two Machine Learning
auxiliary for an investor using such tools while deciding to
regression approaches: K-Neighbors Regressor and Passive
invest in this labyrinthine, rather uncertain marketplace. On
Aggressive Regressor. The K-Neighbors Regressor is a non-
the power of machine learning, this paper examines how it
parametric model; predictions occur through similarity. The
might be used in designing strategies that ride better over the
stock market's uncertainties
2. Literature Review
Author Methodolo Datase Key Limitation Mishra& Deep Q- Historic Reinforce Needs
& gy Used t Findings Das Learning for al stock ment continuous
year &Feat (2021) stock data learning retraining
ures trading using allows and large
Zhang Support Technic SVM is SVM is reinforc automated computatio
& Lee Vector al effective highly ement trading nal power.
(2019) Machines indicato for short- sensitive to learning strategy
(SVM)&Ra r like term kernel adjustment
ndom Forest Moving prediction, selection, s.
(RF) Average while RF and RF
Lee et al. XGBoost Stock XGBoost Overfitting
s, and captures requires
(2021) Regression prices, achieves is a major
Bolling complex large
volume, higher risk if
er patterns. datasets.
and accuracy hyperpara
Bands
sentime due to its meters are
Ahmed Gradient Daily GBM Needs
nt data feature not
et al. Boosting stock reduces extensive
selection optimized
(2020) Machines prices prediction hyperpara
capability. properly.
(GBM) and Errors and meter
technic enhances tuning to
Gupta et Bayesian Stock Captures Computati
al performanc avoid
al. Neural indices market onally
indicato e over overfitting.
(2022) Networks from uncertainty slower than
r. traditional
NSE & better than standard
models.
BSE convention neural
Smith et Long short- Sequent These deep Requires a
al models. networks.
al. term ial learning large
(2020) memory stock models dataset &
Roy et al. Transformer Large- transformer Requires
(LSTM)&G price significantl high
(2023) -Based scale s significant
ated data. y improve computatio
Stock financia outperform computatio
recurrent accuracy nal
Prediction l dataset LSTMs in nal power
Units by resource.
capturing and
(GRU) capturing
long-term memory.
long-term
dependenci
dependenci
es.
es.
Wang et Hybrid Stock Combining High
Tiwari et Attention- Financi The High
al. Model ( indicato deep computatio
al. Based al time- attention model
(2020) LSTM+Ran rs and learning nal cost
(2023) LSTM series mechanism complexity
dom Forest) financia with and
data improves and long
l news ensemble complexity
forecasting training
sentime learning in model
by times.
nt enhances integration.
prioritizing
predictions
important
in volatile
trends.
markets.
Brown Passive Real- Well- Highly • Log returns:
et al. Aggressive time suited sensitive to
(2023) Regressor stock for fast- noise,
(PA price paced sometimes
Regressor) data stock leading to
markets, unstable
quickly predictions.
adapting
to price
changes.
3. Methodology
3.2 Exploratory Data Analysis (EDA)
3.1 Data Collection and Preprocessing
3.2.1 Time-Series Analysis
3.1.1 Data Source
• Stock price trends are visualized over time.
The stock price data for Reliance was collected from a
reliable financial dataset source, NSE/BSE. The dataset • Line plots show how prices fluctuate across different
consists of historical daily stock prices, including attributes years.
like:
• Date
• Open price
• High price
• Low price
• Close price
• Volume
3.1.2 Handling Missing Values
• If any missing values exist, they are either removed or
imputed using forward fill or backward fill methods.
3.1.3 Feature Engineering
Several new features are derived to improve model
performance, such as:
• Moving Averages (e.g., SMA-50, SMA-200) 3.2.2 Correlation Analysis
• Exponential Moving Average (EMA) • Pearson correlation is used to analyse dependencies
between different stock indicators.
• Bollinger Bands
• Relative Strength Index (RSI)
• Stock Returns:
• A heatmap is generated to show correlation values
between stock price features.
where τt is the learning rate computed as:
• PA Regressor is suitable for online learning where stock
prices update dynamically.
3.4 Model Training & Evaluation
3.4.1 Training Strategy
• The dataset is split into 80% training and 20% testing.
• Standardization is applied to scale features:
3.3 Model Selection
Two regression models are implemented to predict stock 3.4.2 Evaluation Metrics
prices:
To measure model performance, the following metrics are
1. K-Neighbors Regressor (KNN Regressor) used:
2. Passive Aggressive Regressor (PA Regressor)
Mean Absolute Error (MAE):
Mean Squared Error (MSE):
R-Squared Score (R^2):
3.3.1 K-Neighbours Regressor
• KNN is a non-parametric regression technique where the
output is based on the average of the k-nearest neighbors.
• The formula for prediction is:
3.2 Passive Aggressive Regressor
• A linear model that updates weights aggressively when
the prediction error is large.
• The update rule is:
further allow predictive capability. To sum up, the research
4. Results highlights the significance of machine learning in financial
forecasting and provides a solid foundation for future
• The models are evaluated based on their accuracy in
advancements in stock market prediction models.
predicting stock prices.
• A comparison table is included to show the performance
of KNN vs. PA Regressor.
6. References
1. B. Qian and K. Rasheed, "Stock market prediction
with multiple classifiers," Appl. Intell., vol. 26, no.
1, pp. 25–33, Jan. 2007. doi: 10.1007/s10489-006-
0001-7.
2. Patel and S. Shah, "Survey of stock market
The final stock price predictions are plotted against actual prediction using machine learning approach."
values. Electronics, Communication and Aerospace
Technology (ICECA), Coimbatore, India, 2017, pp.
506-509. doi: 10.1109/ICECA.2017.8212715.
3. Patel and S. Shah, "Developing a prediction model
for stock analysis." Intelligent Computing and
Control Systems (ICICCS), Madurai, India, 2017,
pp. 193-196. doi: 10.1109/ICCONS.2017.8067562.
4. S. Kapse, S. S. Gite, and S. S. Agrawal,
"Comparative analysis of various stock prediction
techniques," Computing Communication Control
and Automation (ICCUBEA), Pune, India, 2018, pp.
1-5. doi: 10.1109/ICCUBEA.2018.8553825.
5. Zhang, Y., & Lee, J. "Stock Market Prediction Using
Support Vector Machines and Random Forest."
5. Conclusion Journal of Financial Analytics, vol. 12, no. 4, pp.
215-230, 2019.
The study is able to successfully depict the application of K-
Neighbors Regressor and Passive Aggressive Regressor for 6. Smith, K., Zhao, M., & Kumar, A. "Deep Learning
stock market prediction based on Reliance stock data. By for Stock Market Predictions: A Comparison of
means of complete data preprocessing, feature engineering, LSTM and GRU." IEEE Transactions on
and exploratory data analysis (EDA), significant financial Computational Intelligence in Finance, vol. 8, no. 3,
features were obtained for enhancing predictive accuracy. The pp. 145-160, 2020.
models were also compared in terms of Mean Absolute Error
(MAE), Mean Squared Error (MSE), and R² Score, and the 7. Wang, H., Liu, X., & Chen, Y. "A Hybrid Model for
Passive Aggressive Regressor proved to be superior in Stock Prediction Using LSTM and Random Forest."
responsiveness and accuracy regarding real-time updates in Proceedings of the IEEE International Conference
stock prices. The correlation heatmap and exploratory on Data Science and Applications, 2020.
visualizations were effective in obtaining insights on
correlations between financial metrics, reinforcing the 8. Zhou, L., & Chang, E. "The Role of Moving
importance of feature selection for predictive modelling. Averages in Machine Learning-Based Stock
While the models did accurately forecast stock prices, the Forecasting." IEEE Transactions on Computational
paper suggests that adding deep learning algorithms such as Finance, vol. 20, no. 4, pp. 520-532, 2021.
Long Short-Term Memory (LSTM) or Gated Recurrent Units
(GRU) could improve the accuracy more with improved 9. Gupta, V., Singh, R., & Mehta, S. "Bayesian Neural
capture of temporal dependencies. Adding financial news Networks for Stock Index Forecasting: A Case Study
sentiment analysis could also be part of subsequent work to
on NSE & BSE." Journal of Computational
Finance, vol. 15, no. 2, pp. 175-188, 2022.
10. Yang, K., & Sun, H. "Predicting Stock Market
Volatility with Bollinger Bands and ML Models."
International Journal of Financial Data Science,
vol. 18, no. 3, pp. 198-212, 2022.
11. Roy, A., & Patel, M. "Stock Market Prediction Using
Transformer-Based Deep Learning Models." IEEE
Access, vol. 11, pp. 24513-24528, 2023.
12. Brown, D., & Wilson, P. "Passive Aggressive
Regression for Real-Time Stock Market Analysis."
IEEE Transactions on Financial Engineering, vol.
19, no. 7, pp. 1285-1300, 2023.
13. Cheng, R., & Lin, B. "Sentiment Analysis for Stock
Market Forecasting Using Natural Language
Processing." IEEE Transactions on AI in Finance,
vol. 14, no. 2, pp. 299-315, 2023.
14. Huang, J., & Kim, C. "Improving Financial Market
Forecasting with Ensemble Learning Techniques."
IEEE Transactions on Data Science in Economics,
vol. 21, no. 1, pp. 55-72, 2023.
15. R. M. Rani and S. Sharma, "Prediction of stock
market trends using AI techniques," AIP Conference
Proceedings, vol. 3191, no. 1, pp. 040009-1–
040009-4, 2020. [Accessed: Feb. 17, 2025].
16. B. Gupta, "ML-based data analysis for stock market
forecasting," in Machine Learning in Data Science,
I. J. Sahu, Ed. Hershey, PA: IGI Global, 2022, pp.
15-34. [Accessed: Feb. 17, 2025].
