Deep Learning-Based Hybrid Model with Multi-Head Attention for Multi-Horizon Stock Price Prediction
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
The prediction of stock prices is challenging due to their volatility, irregular patterns, and complex time-series structure. Reliably forecasting stock market data plays a crucial role in minimizing financial risk and optimizing investment strategies. However, traditional models often struggle to capture temporal dependencies and extract relevant features from noisy inputs, which limits their predictive performance. To improve this, we developed an enhanced recursive feature elimination (RFE) method that blends the importance of impurity-based features from random forest and gradient boosting models with Kendall tau correlation analysis, and we applied SHapley Additive exPlanations (SHAP) analysis to externally validate the reliability of the selected features. This approach leads to more consistent and reliable feature selection for short-term stock prediction over 1-, 3-, and 7-day intervals. The proposed deep learning (DL) architecture integrates a temporal convolutional network (TCN) for long-term pattern recognition, a gated recurrent unit (GRU) for sequence capture, and multi-head attention (MHA) for focusing on critical information, thereby achieving superior predictive performance. We evaluate the proposed approach using daily stock price data from three leading companies—HDFC Bank, Tata Consultancy Services (TCS), and Tesla—and two major stock indices: Nifty 50 and S&P 500. The performance of our model is compared against five benchmark models: temporal convolutional network (TCN), long short-term memory (LSTM), GRU, Bidirectional GRU, and a hybrid TCN–GRU model. Our method consistently shows lower error rates and higher predictive accuracy across all datasets, as measured by four commonly used performance metrics.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.005 | 0.003 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it