Stock Price forecasting using PSO-trained neural networks

This paper discusses the performance an artificial neural network (ANN) utilizing particle swarm optimization (PSO), to forecast the Singapore stock market index. The particle swarm optimized feed forward neural network (PSO FFNN) program which was developed in C++ will also be discussed. The Straits Times Index (STI) is the primary time series data set and the California electricity market price data will be used as a secondary data set to validate the results obtained from the STI data set. An initial overview of the results obtained from the back propagation neural network (BPNN) optimized parameters will be discussed and used as a benchmark for the PSO FFNN. Subsequently, the improvement in forecasting accuracy after replacing the traditional back- propagation algorithm with particle swarm optimization (PSO) will be shown. Finally, the performance of the PSO FFNN is evaluated by optimizing the PSO parameters and the results are illustrated to show the success of implementation of the particle swarm algorithm in the training of neural network weights.