Comparing the Feature Selection Using the Distributed Non-overlap Area Measurement Method with Principal Component Analysis

Abstract

This paper compares the forecasting performance of the feature extraction using the principal component analysis (PCA) that is one of the oldest and best known techniques in multivariate analysis with the feature selection using the non-overlap area distribution measurement method based on the neural network with weighted fuzzy membership functions (NEWFM). This paper proposes CPPn,m (Current Price Position of day n : a percentage of the difference between the price of day n and the moving average of the past m days from day n-1) as a new technical indicator. In this paper, two and one input features with the best average forecasting performance are selected from the number of approximations and detail coefficients made by Haar wavelet function from CPPn,5 to CPPn-31,5 using the non-overlap area distribution measurement method and PCA, respectively. The performance results of the non-overlap area distribution measurement method and PCA are 60.93% and 56.63%, respectively. The non-overlap area distribution measurement method outperforms PCA by 4.3% for the holdout sets.