TriFusionNode: A Hybrid Attention–Ensemble Framework for Multi-Task Smartphone Rating and Price Prediction
Students & Supervisors
Student Authors
Supervisors
Abstract
Precisely predicting smartphone ratings and prices is a complex task due to the various factors associated with both technical specifications and user features. Although classical machine learning models such as Random Forest, XGBoost, and LightGBM can capture the non-linear relationships among various factors involved, they may not have the representational depth to effectively capture and learn the underlying patterns. Traditional deep neural networks often struggle on heterogeneous tabular data due to overfitting risks and limited interpretability. This study proposes TriFusionNode (TFN), a hybrid multi-task framework that combines attention-based neural feature learning with NODE-motivated residual blocks and ensemble tree-based regressors to jointly predict smartphone ratings and prices. The hybrid framework benefits from feature gating and facet related multi-head attention, combined with uncertainty weighted losses to further enhance stability and generalizability of predictions across the observed application. Based on the experimental results on a real-world smartphone dataset, the proposed architecture of TFN produced better overall prediction performance relative to baseline models, achieved on average lower MAE and RMSE results and higher overall R2 scores. These results highlight the effectiveness of combining continuous- depth-inspired feature transformations with ensemble learning for structured multi-output regression.
Keywords
Publication Details
- Type of Publication:
- Conference Name: International Conference on Electrical, Computer Telecommunication Engineering (ICECTE 2026)
- Date of Conference: 29/01/2026 - 29/01/2026
- Venue: RUET, Rajshahi
- Organizer: Faculty of ECE, Ruet