The Predictability of Fractional Difference Feature using Support Vector Machines

  • ดิฐวัฒน์ ไทยรุ่งโรจน์
  • ธฤตพน อู่สวัสดิ์
Keywords: Support Vector Machines, Machine Learning, Fractional Differencing


The objective of this study is to investigate the predictability of the fractional difference feature using support vector machines. Our sample is the close price of the Stock Exchange Thailand Index(SET Index) from 1 January 2008 to 31 December 2019. We calculate daily return and fractional difference of SET Index and use them as features. The model will use to predict the trend of SET Index for 1, 5, 20, 60 days. The results of unit root test shows that daily return is stationary and fractional difference stationary at d = 0.60

The experimental results show that average accuracy from 5-fold Cross-Validation of daily return is higher than fractional difference feature for 1-day trend prediction which gives average accuracy equal to 53.21%. Fractional difference feature has higher accuracy in trend prediction for 5, 20, 60 days which gives an average accuracy equal to 53.14% 57.01% 60.92% in order. The results demonstrate that fractional difference can improve the predictability of support vector machines for long-term prediction.


Most read articles by the same author(s)

1 2 3 > >>