BSE 30 Financial Forecasting Using Regression Techniques

Abstract

Forecasting financial markets is particularly difficult due to the system's inherent non-linearity and unpredictability. Traditional methods for creating stock market predictive models have several limitations, leading to the widespread adoption of intelligent techniques. This study explores the application of neural network–based regression models utilizing various technical indicators recommended by researchers. Stock index data, including Open, High, Low, and Close values, are used to derive features, which are then selected using numerous existing rank-based feature selection methods. A new index data set with a reduced feature subset is then provided to the neural network–based regression models for predicting future values. Empirical results indicate that feature extraction followed by feature selection can significantly enhance the efficiency of the ANN model. For the experimental study, one years of historical daily Indian stock data were analyzed, From the original data, ten technical-indicator-based features were initially extracted. A rank-based feature selection technique was then applied, resulting in a reduced subset of five optimal features. Using this reduced feature set, the proposed neural network–based regression models achieved a Mean Absolute Percentage Error (MAPE) of 0.594300 on the test dataset.