Life Expectancy Prediction using Explainable AI

Predicting life expectancy accurately serves as a necessity for developing effective public health policies and planning strategies. Machine learning (ML) solutions are more effective than statistical ones in terms of predictive performance but the question of explainability has not yet been clarified. The code will simulate a comprehensive framework, which combines five advanced regression models, such as RF, Xgboost, LightGBM and Cat-Boost with Gradient Boosting to predict life expectancy on a publicly available WHO-based dataset. Mean Absolute error (MAE) as well as R2 score was computed to compare and evaluate the performance of each model. To address the timely issue of model explanability, we incorporate three Explainable AI (XAI) algorithms, i.e., SHAP (SHapley Additive ex-Planations), LIME (Local Interpretable Mod-el-Agnostic Explanations) and LOFO (Leave-One-Feature-Out Importance). RF performed best, its per-performance was 1.03 and its R-Squared was 0.97. HIV/AIDS death rate, adult mortality, schooling and healthcare expenditure were the most significant ones among the feature importance analysis as per the XAI techniques. Relative comparison with the previous researches concerning the same data reveals that this type of analysis can predict better, yet it is more open and comprehensible. The suggested model enables making informed choices in the domain of public health since it leads to high model accu-racy and viable insights.