A Novel LSTM-TCN Deep Learning Model for Stock Price Forecasting in Indonesia

Azkario Rizky Pratama (Universitas Gadjah Mada)
Bagus Rakadyanto Oktavianto Putra (Universitas Gadjah Mada)
Ascariena Rafinda (Sekolah Tinggi Ilmu Manajemen YKPN, Yogyakarta)

JURNAL SISTEM INFORMASI BISNIS
Vol 16, No 1 (2026): Volume 16 Number 1 Year 2026 (In Press)

Making predictions about stock prices is challenging since financial markets are always changing and dynamic. Our study involves analyzing how Long Short-Term Memory (LSTM), Temporal Convolutional Network (TCN) and their combination work in predicting stock prices. We obtained raw data from Yahoo Finance on the stock prices of five major banks and then prepared it for our analysis through imputation, normalization and using a sliding window. The Adam optimizer was used to fine-tune the learning rates when training the baseline models LSTM and TCN. According to our results, TCN performs better than LSTM when forecasting the prices of stocks. However, using LSTM-TCN as a hybrid model results in the lowest errors among all metrics, including MSE and MAE. It is demonstrated that combining LSTM with parallel TCNs improves the accuracy of predictions. From the study, it can be seen that the LSTM-TCN hybrid model could be a valuable tool for forecasting stock prices.