How effectively can AI predict stock prices across different markets, and which factors contribute most to prediction accuracy?

Abstract

This thesis examines how machine learning methods can be applied for predicting stock market and provides an analysis of these techniques. Machine learning has gained recognition to model complex patterns in financial time series with more data and computing power. This investigation assists in capturing relationships in financial data and reviews research from 2010 to 2025 to examine how efficiently different machine learning methods perform. The work reviews both simple methods such as random forests and complex deep learning networks utilizing real data from several investigations and integrates machine learning forecasting with established economic theories such as the Efficient Market Hypothesis, Behavioural Finance, and Adaptive Markets Hypothesis to place these capabilities within broader theoretical frameworks. This investigation integrates machine learning methods with famous economic concepts and challenges traditional market efficiency theories by analyzing how machine learning models can detect market inefficiencies caused by changing market conditions and human behavioural biases. Moreover this research investigates how these models can detect opportunities that arise from the market's inability to correct itself completely. Problems regarding algorithmic transparency, fairness, and regulatory compliance in machine learning financial applications have been addressed in this thesis.

In addition this research examines the ethical and sustainable effects of employing machine learning in the stock market and utilizes systematic review methods to gather and combine performance data from multiple studies. Meta-analysis techniques have been implemented to merge the results together and the right techniques were applied for preparing data, testing models, and avoiding the common issues such as overfitting and data leakage. The results suggest that there are gaps in existing studies about situations where machine learning models perform better compared to traditional econometric methods. This study suggests ideas for future study directions and provides valuable information to academic researchers and financial practitioners regarding the current state of machine learning based financial forecasting. The research provides details about recent machine learning methods in stock prediction and gives direction for building trustworthy machine learning models and emphasizes the need for continuous evaluation and updates in changing market conditions.