Development and evaluation of novel forecasting adaptive ensemble model

This paper proposes a new ensemble based adaptive forecasting structure for efficient different interval days' ahead prediction of five different asset values (NAV). In this approach three individual adaptive structures such as adaptive moving average (AMA), adaptive auto regressive moving average (AARMA) and feedback radial basis function network (FRBF) are employed to first train with conventional LMS, conventional forward-backward LMS and corresponding learning algorithm of FRBF respectively. After successful validation of each model the output obtained by each individual model is optimally weighted using Genetic algorithm (GA) as well as particle swarm optimization (PSO) based techniques to produce the best possible different days ahead prediction accuracy. Finally the results of prediction obtained of the NAV values are compared with the results obtained by individual predictors as well as by other four existing ensemble schemes. It is in general demonstrated that in all cases the proposed forecasting scheme outperforms other competitive methods.