Abstract
Stock is one type of long-term investment that is quite in demand by the public because this investment brings quite a large profit for its investors. However, in relation to this, stock price movements, in general, tend to be non-linear and non-stationary, this is because stock prices can be influenced by several factors whose results can change the pattern of stock price values either up or down, so this can make it difficult to stock prices prediction. In this study, a comparative analysis of prediction models was carried out in predicting stock prices using a technical approach based on past data, while the data used were historical stock prices by taking data samples from three issuers from the Indonesian capital market. There are three methods that were tested in this study, including Linear Regression (LR), Random Forest Regression (RFR), and Multilayer Perceptron (MLP). The test was carried out with two data modeling, namely partitioning which was validated with Cross-Validation, and data modeling with Cross-Validation without partitioning. In this study, the prediction model with LR is able to produce a fairly low error prediction value with the lowest RMSE score of 0.010 and the highest RMSE of 0.012, the lowest MAPE of 1.2%, and the highest of 1.9%, the lowest MAE of 0.006 and the highest. of 0.009, and the highest R2 value was 99.8% and the lowest was 99.6%. It can be concluded that in this study, the Linear Regression prediction model is able to predict historical data on stock prices better than the RFR and MLP models.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have