Comparative analysis of computational intelligence techniques in financial forecasting: A case study on ANN and ANFIS models

Abstract

This paper explores the evolution of financial analysis and forecasting models, contrasting traditional statistical methods with computational intelligence techniques. While conventional methods like autoregressive moving average and exponential smoothing struggle with the complexity of financial time series, computational techniques offer more effective modeling and prediction. Notably, ANN models excel in forecasting significant price movements based on past data, promising a more systematic approach to predicting future prices. However, selecting the right modeling techniques remains critical, considering the strengths and weaknesses of each method. Comparative analysis between ANN and ANFIS reveals their distinct advantages and disadvantages, guiding method selection in financial forecasting. While ANN handles large datasets and complex relationships well, ANFIS offers flexibility and noise handling capabilities. Despite both methods having drawbacks like overfitting and high computational demands, this analysis aids in method selection. Through evaluation using RMSE, MAPE, and R2 metrics, this study provides quantitative measures for assessing model accuracy and performance. Additionally, employing a correlation-based feature selection strategy enhances model efficiency, highlighting the importance of dataset reduction in improving model performance and interpretability.