Explainable Pediatric Lung Capacity Prediction Using Category Embeddings and Mixture Density Networks
2nd International Symposium on AI-Driven Engineering Systems, ISADES 2026, Hybrid, Mbale, Uganda, 19 - 20 Haziran 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/isades69945.2026.11608125
- Basıldığı Şehir: Hybrid, Mbale
- Basıldığı Ülke: Uganda
- Anahtar Kelimeler: Categorical Embedding Model, Deep Learning, Explainable Artificial Intelligence (XAI), LIME, Lung Capacity Prediction, Mixture Density Networks, Probabilistic Regression, SHAP
- Hacettepe Üniversitesi Adresli: Evet
Özet
This study presents an explainable probabilistic deep learning framework for pediatric lung capacity estimation by using tabular clinical data. The proposed approach merges a Category Embedding Model with a Mixture Density Network (MDN) head for modeling the complex nonlinear relationships between input variables and lung capacity while also capturing predictive uncertainty. Within the proposed network, categorical variables are firstly transformed into embedding feature repre-sentations for improving feature interaction learning, while the MDN head predicts conditional probability distributions instead of deterministic point predictions, which enables uncertainty-aware regression analysis. Moreover, two explainable artificial intelligence (XAI) techniques, SHAP and LIME, are integrated into the framework for providing both global and local explain-ability of model decisions and feature contributions. Experimental analyses show that the proposed model effectively captures the underlying relationships affecting pediatric lung capacity and provides interpretable predictions with enhanced modeling flexibility for tabular healthcare data.