Prediction of stock purchase decisions using artificial neural network method

Abstract

The difficulty of determining a stock purchase decision is a problem to benefit from stock transactions. This study aims to give a person's decision to buy one of the issuer's company shares to get a profit on the same day. The dataset used in this study came from the investing.com website in the form of daily data shares of PT Indofood CBP Sukses Makmur Tbk with ICBP stock code for the period January - September 2019. The attributes used in this study were the opening price, highest price, lowest price, closing price, transaction volume, day representation, and decisions. The dataset that has been collected was normalized using the Min-Max method to facilitate data processing. This research used the backpropagation neural network method and used the 10-Fold Cross Validation and Confusion Matrix for validation. The results of this study indicate that the backpropagation neural network method uses the bipolar activation function with training cycles of 2000 and learning rate of 0.03 has the best performance namely 69.35% of accuracy, 67.65% of precision, 74.19% of recall and 30.65% of error rate for prediction of stock purchase decisions per day in the form of buy or not.