Skip to content
Open access

Forecasting Next-Day TLKM Stock Prices Using Linear and Nonlinear Machine Learning Models: A Case Study of PT Telekomunikasi Indonesia Tbk

Oct 2026 · JURNAL TEKNIK INFORMATIKA · 0 citations · 21 references

Abstract

Stock price forecasting provides useful information for investors in making decisions in fluctuating capital markets. The objective of this work is to develop and compare models to forecast the closing price for the next trading day. Three models were employed: linear regression, an optimized random forest regression, and an optimized backpropagation neural network. The dataset used consists of daily market data for TLKM and IHSG covering the period from March 22, 2018, to February 24, 2027. The predictors used include open price, high price, low price, trading volume, SMA5, and RSI14, while the target variable is the next day's TLKM closing price. The dataset was divided chronologically into 80% training and 20% testing sets. Time-series cross-validation was performed using 5-fold TimeSeriesSplit. Hyperparameter optimization was applied to the random forest and backpropagation neural network models using RandomizedSearchCV. The results show that linear regression outperforms the other models based on the overall evaluation metrics, with an MAE of 57.67, RMSE of 81.38, MAPE of 1.37%, and  of 92.8%. The MAPE value indicates that the linear regression model has a relatively low forecasting error. The linear regression model can also explain approximately 92.8% of the variation in the next day's TLKM closing price through the predictors included in the model. The implication of this work is that a linear regression model can provide an additional quantitative reference for investors in evaluating the possibility of TLKM closing price movements on the next trading day.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.