CLASSIFICATION OF STOCK PRICE MOVEMENTS USING MULTILAYER PERCEPTRON WITH TREND, MOMENTUM, AND VOLUME INDICATORS

Abstract

ABSTRACT

 Stock trading focuses on leveraging fluctuations in stock prices to make a profit. Trading strategies are developed to help determine the right moment to buy and sell stocks in the market. These strategies can be built manually through fundamental, technical, and sentiment analysis, or using Machine Learning techniques that process historical stock data to generate buy and sell signals, enabling easier decision- making. This research proposes a Machine Learning model using the Multilayer Perceptron Backpropagation algorithm with four layers, which utilizes technical indicators as features. The model uses seven main features, namely: momentum value, Price Oscillator, difference between On-Balance Volume (OBV) value and the average OBV of the preceding five days, change in distance between price and Moving Average (MA) 10, distance between Relative Strength Index (RSI) value and MA10 (RSI), change in distance between MA20 and MA50, and difference between %K and %D of Stochastic indicator. The classification categories are divided into three classes, namely: Buy, Sell, and Hold. The historical data used includes closing price and volume for 10 years, from January 1, 2014 to December 31, 2023, taken from Yahoo! Finance. The model is optimized using a Risk-Reward Ratio of 1:2, with a profit target of 6% and a loss limit of 3%, and an evaluation period of 10 days. Model testing was conducted on ASII, TLKM, PWON, BBRI, BBCA, BNGA, UNVR, GGRM, and HMSP stocks, which represent categories of stocks with sideways, uptrend, and downtrend trends. The test results provide an average accuracy rate of 65%, with the greatest accuracy in BBCA stock at 75%, and the smallest accuracy in PWON stock at 57%. From the confusion matrix, the average win-loss ratio value is 5.23 times. This indicates that the success rate of transactions that generate profits is 5 times that of transactions that generate losses.

Keywords: stock price, stock trading, machine learning, technical indicators.