Stock closing price prediction based on combined model of PCA-IMKNN

Abstract

This paper proposes an improved multidimensional k-nearest neighbours (IMKNN) method based on principal component analysis (PCA), which uses multi-dimensional time series to predict the closing price of stocks. The IMKNN method is proposed based on MKNN. Compared with MKNN, its distance value has more possibilities, which can overcome the difficulty of sorting caused by too limited possibilities of distance values in the existing MKNN method. The newly proposed PCA-IMKNN method utilises the advantages of PCA and IMKNN and has high prediction accuracy for short-term prediction. Four representative stocks (NAS, S&P500, DJI, and RUT stock index) and four general evaluation criteria were used to test the method. The results show that, compared with the MKNN method, IMKNN method, and PCA-MKNN method, the proposed PCA-IMKNN model has higher prediction accuracy.