MODELING TIME SERIES FORECASTING USING EVOLUTIONARY TECHNIQUES

Abstract

In recent years, usage of time series forecasting has been increasing day by day for prediction like, share market, weather forecasting and data analysis. Forecasting of Mackey Glass chaotic time series has been carried out in this paper. It is considered that prediction of a chaotic time series system is a nonlinear, multi-variable and multimodal optimization problem. To get an optimum output of times series, global optimization techniques are required in order to minimize the effect of local optima. Application of recent evolutionary techniques have been considered as pervasive technology for Optimization. In this paper, Fuzzy Logic System (FLS) deals with nonlinearity and generates the rule base from training data used for time series forecasting. Further, application of five recent evolutionary techniques have been considered for optimization like Genetic Algorithm (GA) and Gravitational Search Algorithm Particle Swarm Optimization (GSA-PSO),. A comparison for bench mark data of time series forecasting is done using above discussed techniques. it is observed that GA performs better as compared to GSAPSO in both terms, i.e. accuracy and time