PREDIKSI DATA HARGA SAHAM HARIAN MENGGUNAKAN FEED FORWARD NEURAL NETWORKS (FFNN) DENGAN PELATIHAN ALGORITMA GENETIKA (Studi Kasus pada Harga Saham Harian PT. XL Axiata Tbk)

Abstract

Artificial neural network (ANN) is an information processing system that has similar characteristics with biological neural networks. ANN consists of neurons arranged in layers and has a pattern of connectedness within and between the layers called the network architecture. Feed Forward Neural Networks model (FFNN) is the NN model that have a fairly simple network architecture with one hidden layer and can be applied for the prediction of time series data. In general, FFNN trained using Backpropagation algorithm to obtain weights. Backpropagation can work well on a simple training issue but performance will decrease and trapped in a local minimum when applied to data that have great complexity. The solution to this problem is to train FFNN using Genetic Algorithm (GA). GA is a search algorithm that is based on the mechanism of natural selection and genetics to determine the global optimum. Prediction of time series data using conventional methods such as Autoregressive Integrated Moving Average (ARIMA) is limited by the presence of assumptions. The existence of the assumptions that must be met in using the ARIMA model shows weaknesses of the model to be used as a predictive tool especially for financial data such as stock price data that tend to have complicated patterns. These conditions encourage to try to use the model training FFNN with Genetic Algorithm for prediction of stock price data, but the problem is how to understand the workings of FFNN training using the GA, the determination of the combination crossover probability (𝑝 ), the number of populations, the number of generations, and the tournament size (k) to produce predictive value which approaching actual value. One possible option is to use the technique of trial-end-error by experimenting for some combination of these four parameters. Of the 64 times application of the AG to train FFNN model to PT. XL Axiata’s daily stock price obtained results are sufficiently accurate predictions indicated by the proximity of the target to the 𝑐 output with the crossover probability (𝑝 ) 0.8, the number of populations 50, the number of generations 20000 and tournament size 4 produces the testing RMSE 107.4769. 𝑐 Keywords : prediction of daily stock price data , neural networks , feed forward neural network , genetic algorithm