Research Article
Comparative Analysis of Ensemble LSTM and Random Forest for Stock Trend Prediction: A Many-to-One Transfer Learning Approach
@INPROCEEDINGS{10.4108/eai.22-5-2026.2365140, author={Pengtian Si}, title={Comparative Analysis of Ensemble LSTM and Random Forest for Stock Trend Prediction: A Many-to-One Transfer Learning Approach}, proceedings={Proceedings of the 4th International Conference on Image, Algorithms, and Artificial Intelligence, ICIAAI 2026, 22-24 May 2026, Singapore, Singapore}, publisher={EAI}, proceedings_a={ICIAAI}, year={2026}, month={8}, keywords={Stock prediction LSTM Random forest Transfer learning Financial time series}, doi={10.4108/eai.22-5-2026.2365140} }- Pengtian Si
Year: 2026
Comparative Analysis of Ensemble LSTM and Random Forest for Stock Trend Prediction: A Many-to-One Transfer Learning Approach
ICIAAI
EAI
DOI: 10.4108/eai.22-5-2026.2365140
Abstract
Discovering patterns in the diverse financial time series is an extremely challenging task. The Efficient Market Hypothesis holds that such efforts are futile, but the latest advancements in the field of deep learning have reignited this debate. This paper explores a "many-to-one" transfer learning strategy. This paper uses a dataset to compare the robust tree-based logic of the Random Forest (RF) regressor with the time-sensitive network of the Long Short-Term Memory (LSTM) network. The empirical findings on the data of 2024 indicate that there is a strong difference, yet Ensemble LSTM has done well in pattern recognition, which the direction accuracy rate of 54.35%, the RF model surprisingly has a better level of strategy return of 40.11%, as compared to the benchmark of 37.06%. The ultimate conclusions reveal that it is possible that the tree-based models have aggressive robustness in returning high volatility profit opportunities.
