Foreign Exchange Rate Prediction using Hybrid of ANFIS and Wavelet with Feature Extraction and Feature Selection

Abstract

Foreign exchange rate prediction is required for making strategy of foreign trading and other financial purposes. The ability to accurately predict the future behavior of time series data is very crucial using statistical methods; instead, hybrid techniques may perform better than individual.  This paper focuses on hybrid approach of three different techniques: wavelet for removing noise and non-linearity from chaotic time series data, feature extraction for extracting new features and feature selection for selecting best features. Adaptive Neuro-Fuzzy Inference System (ANFIS) is applied for prediction of linear Foreign Exchange (FX) data. The empirical result shows that hybrid approach with optimized extracted features: SMA (Simple Moving Average), WMA (Weighted Moving Average) and VAR(Variance) provides the best predictive result with Mean Absolute Percentage Error (MAPE)=1.246, Mean Absolute Error  (MAE)= 0.011 and Root Mean Square Error (RMSE)= 0.013.