A Review of Soft Computing Techniques for Time Series Forecasting

Abstract

Objectives: This paper gives a brief survey analysis on time series forecasting models developed so far and how soft computing techniques are becoming popular for time series forecasting. Statistical Analysis: This paper gives an introductory study on time series models develop in literature so far. First all traditional models were developed among them ARIMA became popular but only for linear data. Then after Soft Computing models came into picture like ANN and it was proved to be best for nonlinear forecasting. But none of the individual models are capable to handle all types of datasets as real data is a mixture of linear and nonlinear both. So there arises a need for hybridization. Findings: Many traditional methods have been developed since last few decades for time series forecasting, however their performance is not up to the mark till today. Recent trends have proven that soft computing techniques like neural network, support vector machine are good alternatives to conventional methods. The accuracy of time series model is very important and difficult task to achieve, and individual models are not able to perform well for all types of time series data. So, hybridization of models is better to achieve good accurate forecasting results. Application/Improvements: Implementation of a hybrid model using soft computing techniques which can give better accurate results as real data are made linear and nonlinear both. This hybrid model can be used in applications like weather forecasting, exchange rate forecasting, etc.