K-Nearest Neighbor Regression with Principal Component Analysis for Financial Time Series Prediction

Abstract

This paper constructs an integrated model called PCA-KNN model for financial time series prediction. Based on a K-Nearest Neighbor (KNN) regression, a Principal Component Analysis (PCA) is applied to reduce redundancy information and data dimensionality. In a PCA-KNN model, the historical data set as input is generated by a sliding window, transformed by PCA to principal components with rich-information, and then input to KNN for prediction. In this paper, we integrate PCA with KNN that can not only reduce the data dimensionality to speed up the calculation of KNN, but also reduce redundancy information while remaining effective information improves the performance of KNN prediction. Two specific PCA-KNN models are tested on historical data sets of EUR/USD exchange rate and Chinese stock index during a 10-year period, achieving the best hit rate of 77.58%.