Hybrid Deep Learning and Time Series Analysis for Stock Market Prediction: A Multilayer Perceptron and ARIMA Modelling Approach

Predicting the stock market can be extremely difficult because of the complicated nature of financial markets. Although ARIMA models are good at capturing linear trends, they have limits when it comes to nonlinear trends. MLP models excel in intricate relationship capture but falter in short-term trend identification. Recognizing these constraints, this paper introduces a ground-breaking hybrid approach, fusing Autoregressive Integrated Moving Average (ARIMA) and Multilayer Perceptron (MLP) models to bolster stock market forecasting. The proposed hybrid model innovatively integrates ARIMA and MLP, concurrently harnessing linear and nonlinear dependencies in stock market data for a more holistic and accurate predictive framework. The methodology entails a sequential application of ARIMA and MLP models. ARIMA addresses short-term linear trends, while MLP concentrates on nonlinear patterns and long-term dependencies. Ensemble techniques amalgamate predictions from both models, forging a synergistic approach that maximizes their complementary strengths. Performance assessment, using historical stock market data and metrics such as Mean Squared Error (MSE) and Mean Absolute Error (MAE) with values 12.67 and 3.12, validates the proposed hybrid model's superiority over individual ARIMA and MLP models. This pioneering hybrid ARIMA-MLP approach mitigates the limitations of existing models, providing a more robust and precise solution for stock market prediction.