Application of Restricted Boltzmann Machine as Feature Extracting in Predicting the Trend of Indonesia Stock Market Price Marshanda Tiana Sarjito (a)
(a) Faculty of Mathematics and Natural Sciences, Bandung Institute of Technology, Indonesia
Abstract
Econophysics is an interdisciplinary science that applies physics in solving economic phenomena. One of the focuses of econophysics is the study of the stock market. Making predictions for the stock market is very important because it can provide financial decisions to investors. Many approaches have been done in predicting stock prices such as statistical models, machine learning, and deep learning. However, the use of high dimensional data such as stock prices allows a reduction in computational performance so that a deep learning model, Restricted Boltzmann Machine (RBM), is proposed as a feature extracting. The purpose of this study is to analyze the effect of RBM in predicting stock price trends and compare Gaussian RBM with Bernoulli RBM. This research was conducted by adding RBM in the prediction of stock price trends carried out for three stock prices namely ASII, BMRI, and TLKM using Support Vector Machine (SVM), Random Forest, Naive Bayes Classifier, and Multi Layer Perceptrons Classifier. The results of the prediction are in the form of accuracy values that are evaluated using matrix evaluation. From this research, it is concluded that the use of RBM has an influence on the resulting accuracy value and the use of Bernoulli RBM is better than Gaussian RBM. Even so, there are some stock data and prediction classification models that do not show the influence of RBM. This can be caused by many factors such as data characteristics, classification model selection, and RBM hyperparameter selection.
