Transformer-Based Multi-Modal Respiratory Disease Classification with Hybrid Feature Selection and Genetic Optimization

Background and Aims: Chronic respiratory diseases, such as Chronic Obstructive Pulmonary Disease (COPD), are influenced by a complex interplay of pulmonary function, nutritional status, and biochemical markers. Traditional machine learning models often fail to capture the intricate relationships across these diverse data modalities. Methods: In this study, we propose a novel Transformer-based model with multi-modal feature fusion to classify individuals into 'Case' (respiratory disease) and 'Control' (healthy) groups. Using a dataset of ۹۰ subjects with ۱۹۰ features, including pulmonary indices (e.g., FEV۱, FVC), nutritional intake (e.g., protein, kilocalories), and biochemical markers (e.g., IL۱B, MDA)—we leverage the Transformer’s attention mechanism to model dependencies across modalities and time points (e.g., repeated measurements). The dataset was preprocessed to handle missing values and normalized for consistency. Results: Our model achieved an accuracy of ۹۲.۸۶%, outperforming baseline models such as Random Forest and Logistic Regression. The attention mechanism highlighted key features, such as FEV۱/FVC and CAT scores, while uncovering novel associations between nutritional deficits and disease severity. Conclusion: This approach demonstrates the potential of Transformer-based architecture for small-scale, multi-modal medical datasets, offering improved interpretability and predictive power. These findings provide a scalable tool for early detection and personalized management of respiratory diseases, paving the way for advanced AI-driven diagnostics in respiratory health research.

کلیدواژه ها:

Transformer ، Model Multi-Modal ، Data Respiratory Disease