The feed forward neural network with genetic algorithm for daily stock prediction

Abstract

The power of Feed Foward Neural Network (FFNN) in conjungtion with Genetic Algorithm (GA) was applied in this research to predict daily stock price. Finance time series data has a high complexity, so that the acurate prediction is hard to be gained by standard model. Machine learning becomes the new prediction tool which is often used because of its adaptive properties. Neural Network (NN) is one of the machine learning which able to complete inference tasks such as prediction, especially in large data sets. FFNN is one of the NN models that has simple network architecture. In the standard optimization method, the initial weights is randomly selected to desire the optimum solution. But It usually raises the problem of unsteady estimation. The GA optimization method was applied in this research to overcome this lack. GA optimizes any function effectively and seeks a global optimum solution efficiently. GA implementation on the FFNN was aimed to obtain optimum weights that minimizing the error. The daily stock price prediction of PT. Adhi Karya Tbk had RMSE of training and testing data at 51.2531 and 44.8706 respectively. This result was equivalent with MAPE values at 1.5714% and 1.5501%.