A dynamic fuzzy membership degree prediction approach to stock time series

EasyChair Preprint 1299

Abstract

Fuzzy time series analysis is the most successful method in enrollment prediction, stock index forecasting, and temperature prediction. It often suffers from high time complexity and low prediction accuracy due to equidistant partition and formulation of fuzzy relationships. In this paper, we propose a concise method called dynamic fuzzy membership degree prediction (DUMP) with four steps for stock time series. First, a number of fuzzy membership degree time series are constructed from the original one. Second, respective prediction models are built with these time series. Third, dynamic fuzzy membership degrees are predicted using these models. Fourth, the final prediction is obtained through the fuzzification of the degree of membership. Comparison study is conducted on 196 stock price time series across one year in comparison with two state-of-the-art approaches. Results show that our approach generally outperforms existing ones in terms of MSE, MAE and MAPE.

Keyphrases: Defuzzification rule, Dynamic membership function, Stock price prediction, Upper and lower limits

@booklet{EasyChair:1299,
  author    = {Zhao Chunlan and Cao Xijuan and Xie Xiangjun},
  title     = {A dynamic fuzzy membership degree prediction approach to stock time series},
  howpublished = {EasyChair Preprint 1299},
  year      = {EasyChair, 2019}}