Forecasting US stock price using hybrid of wavelet transforms and adaptive neuro fuzzy inference system

Artificial Neural Network (ANN) techniques are often used for time-series data forecasting, and Fuzzy Logic (FL) is integrated with the ANN to improve forecasting. This research aims to develop forecasting models using a hybrid approach of ANN and FL as Adaptive Neuro-Fuzzy Inference System (ANFIS) and Wavelet Transform (WT) for two US stock indices DOW30, and the NASDAQ100. The study also investigates the outcome of various types and numbers of fuzzy Membership Functions (MFs) in the forecasting process. All proposed models were analyzed using various performance majors, and it was discovered that the WT-ANFIS models outperformed the original ANFIS models. In addition, an experimental investigation was conducted using three distinct WT filters. The empirical results demonstrate that the model with the trapezoidal membership function (MF) outperforms the model with the bell-shaped MF with the maximum accuracy. The numbers and types of fuzzy MF have clearly been found to play a substantial effect in the forecasting process.