This project sits between deep learning and finance, turning noisy intraday IDX market data into a disciplined forecasting setup. The work focused on feature engineering from ticker returns and volumes, causal sequence modeling with TCN blocks, validation against strong baselines, and practical calibration so the model stayed grounded around small market movements instead of overreacting to noise.
Deep Learning / Finance / 2026
IDX Intraday Stock Return Prediction using TCN
Developed a causal multivariate forecasting model to predict 15-minute future log returns for 100 target stocks on the Indonesia Stock Exchange, using high-dimensional market features, Temporal Convolutional Networks, baseline comparisons, and calibration for near-zero market movement.