A New Self Adaptive Differential Evolution: Its Application in Forecasting the Index of Stock Exchange of Thailand

This paper proposes wDE, a new differential evolution with self-adaptive CR and F, and adaptive NP. Its performance is evaluated on the 25 benchmark problems in CEC2005. The results indicated that wDE is comparative to SaDE in most benchmark problems and has ability of escaping from local optima in some complex problems. Its adaptation of NP helps accelerate the convergence. In a real-world application, wDE is applied to forecast the market index of Stock Exchange of Thailand during a period of January 2003 to September 2006. The forecasting model provides a more accurate prediction than the multiple linear regression over the same data set with statistical significance. In addition, the model does not require fine-tuning of essential parameters in which it takes time and experience.