Robust backpropagation neural network using date palm seed growth algorithm for stock market prediction

Backpropagation Neural Network (BPNN) is one of the most commonly used models for stock market prediction due to its ability as universal estimators. However, the Ordinary Least Squares (OLS)-based training in BPNN leads to nonrobust weightage estimates in the presence of outliers. Consequently, it affects the prediction performance of the BPNN model. Addressing this issue, this study proposes an alternative approach by replacing OLS with Date Palm Seed Growth Least Median Squares (DPSG-LMedS) algorithm. This approach aims to improve the prediction accuracy at different levels of data contamination in stock market. DPSG-LMedS involve five phases which are training the network iteratively by minimizing the median of estimated errors, removing outliers based on robust standard deviation, retraining on the cleaned data, and stopping once the best LMedS errors meet the setting criteria. Next, the model performance is evaluated using simulated and real data. In simulation analysis, the accuracy of the new model is assessed based on different levels of data contamination (0% to 65%), input lags (5 to 45), and hidden node (5 to 45) configurations. Real data of FBM KLCI stock market closing prices is used to compared the performance of the new model with BPNN and BPNN with LMedS. The best-performing model is determined based on the lowest values of Root Mean Square Error (RMSE) and Geometric Root Mean Square Error (GRMSE). Results from simulated analysis shows that the new model performed well at all levels of data contamination with configuration moderate lags input and lowest hidden nodes. Comparison using real data indicate that the new model outperformed other models. This new model offers a more reliable predicting model and is expected to support investors, economists, policymakers, and financial institutions in making more accurate and informed decisions. Additionally, it contributes to the development of robust neural network techniques for financial prediction applications.