MACHINE LEARNING FOR STOCK MARKET SECURITY MEASUREMENT: A COMPARATIVE ANALYSIS OF SUPERVISED, UNSUPERVISED, AND DEEP LEARNING MODELS

Аннотация

INTRODUCTION data accessibility have accelerated the adoption of The stock market has long been a critical machine learning in stock market analysis. Kaggle, component of the global economy, with significant a widely used data platform, hosts numerous high- implications for individuals, corporations, and quality datasets that encompass financial news, governments. However, due to its dynamic and historical stock prices, and company volatile nature, predicting stock market behavior fundamentals, providing a rich foundation for and identifying security risks within it has machine learning research. Studies leveraging remained challenging. Traditionally, market these data resources have demonstrated the analysis relied heavily on statistical methods and effectiveness of machine learning in various human intuition, which, while valuable, often financial tasks, including stock price forecasting, struggled with complex patterns and fast-paced sentiment analysis, and anomaly detection (Rundo data. With the advent of machine learning (ML), a et al., 2019). Machine learning approaches, new horizon has emerged, offering sophisticated including supervised, unsupervised, and deep tools capable of managing large datasets and learning algorithms, allow researchers to examine uncovering complex, non-linear relationships. different facets of market behavior. For example, Machine learning models can process historical supervised learning models, such as Random stock prices, news sentiment, and macroeconomic Forest and Support Vector Machines (SVM), have indicators to provide deeper insights into market been effective in identifying patterns within behavior and potential security risks (Chen et al., historical stock data, while unsupervised models 2019; Patel et al., 2015). like K-Means and DBSCAN excel at detecting Recent advancements in computational power and anomalies and clustering similar data points https://www.theamericanjournals.com/index.php/tajet 64 THE USA JOURNALS THE AMERICAN JOURNAL OF ENGINEERING AND TECHNOLOGY (ISSN – 2689-0984) VOLUME 06 ISSUE11 (Huang et al., 2019; Zhu et al., 2021). classifying stock data, LSTM models often provide superior results for time-series predictions (Jiang A key challenge in stock market analysis is its et al., 2017). Despite the varying strengths of each reliance on various data types, each contributing model, incorporating multiple approaches can distinct insights into market dynamics. For enhance overall system robustness, providing a instance, sentiment analysis on financial news can comprehensive view of stock market security reveal public perception and its impact on stock risks. prices, while technical indicators derived from stock data help in trend forecasting. Feature Given the complex nature of financial markets, engineering, the process of creating meaningful explainability and interpretability have become features from raw data, has proven essential in critical in machine learning applications for stock extracting valuable insights from complex market security measurement. Stakeholders datasets. Features like moving averages, Relative require transparency in model decisions, Strength Index (RSI), and fundamental ratios particularly when large investments and risks are provide machine learning models with a well- involved. Techniques such as SHAP (SHapley rounded dataset, improving the predictive Additive exPlanations) and LIME (Local accuracy of stock trends and anomalies (Fischer & Interpretable Model-Agnostic Explanations) allow Krauss, 2018; Zhang & Li, 2020). practitioners to assess feature importance and understand model behavior in financial contexts. Deep learning, particularly Long Short-Term These tools have helped bridge the gap between Memory (LSTM) networks, has shown remarkable complex machine learning models and actionable potential in time-series analysis for stock market insights, enhancing trust and usability in financial forecasting. LSTM networks are designed to decision-making processes (Lundberg & Lee, manage sequential data, making them well-suited 2017; Ribeiro et al., 2016).machine learning for predicting stock trends over time. Studies have presents a powerful toolkit for stock market shown that LSTM can effectively capture market security measurement, with diverse models trends, outperforming traditional methods and offering unique advantages. This study seeks to simpler machine learning models in sequential implement a comprehensive methodology, data prediction (Siami-Namini et al., 2018). drawing on supervised, unsupervised, and deep However, deep learning models require learning models, to measure stock market security substantial data preprocessing and are through a combination of historical data, computationally intensive, which can limit their sentiment analysis, and financial indicators. By scalability and increase model complexity (Dixon examining model performance and et al., 2020). interpretability, this research aims to contribut