Grey relational with BP_PSO for time series foreasting

This paper proposes an effective hybridization of grey relational analysis (GRA) and Backpropagation Particle Swarm Optimization (BP_PSO) for time series forecasting. The hybridization employs the complementary strength of these two appealing techniques. Additionally the combination of GRA and BP as cooperative feature selection (CFS) has successfully assessed the importance of each input variable and automatically suggest the optimum input numbers for the forecasting task. Therefore it assists the forecaster to choose the optimum number of dominant input factor without a need to acquire expert domain knowledge. It also helps to reduce the interference of irrelevant inputs on the forecasting accuracy performance. To test the effectiveness of the proposed hybrid GRABP_PSO, the dataset of closing price from Kuala Lumpur Stock Exchange (KLSE) is used. The results show that the proposed model, GRBP_PSO out performed BP_PSO model and BP model in term of accuracy and convergence time.