MACHINE LEARNING BASED MODELLING OF TEMPORAL PATTERNS IN FINANCIAL MARKET DATA

Forecasting financial markets is a challenging problem due to their highly volatile, dynamic, and non-linear behavior. Accurate predictions are critical for investors, traders, and policymakers to mitigate risks and optimize decision-making. In this paper, We assess and contrast five machine learning algorithms for predicting the stock prices of Apple Inc. (AAPL): Random Forest, XGBoost, Support Vector Machines (SVM),Long Short-Term Memory (LSTM), and Gated Recurrent Units (GRU). Moving averages and momentum indicators were among the feature engineering techniques used to gather and pre-process historical stock data. Mean Squared Error (MSE), Mean Absolute Error (MAE), and prediction accuracy were used to train and assess each model.Experimental results demonstrate that deep learning approaches, particularly GRU, achieve the highest accuracy (94.3%), effectively capturing long-term temporal dependencies. Conversely, ensemble learning techniques like XGBoost andRandom Forest provide robust performance in handling non-linear patterns, while SVM achieves competitive results with smaller datasets.The findings highlight the advantages of integrating multiple machine learning paradigms and suggest the potential of hybrid forecasting systems that combine traditional ensemble models with deep learning architectures for improved market prediction.