Extraction of human understandable insight from machine learning model for diabetes prediction

Explaining the reason for model’s output as diabetes positive or negative is crucial diagnosis. Because, reasoning predictive outcome of model helps to understand why predicted an instance into class. In recent years, highest accuracy and promising result achieved with simple linear complex deep neural network. However, use such ensemble learning have trade-off between interpretability. response problem interpretability, different approaches been proposed explain model. relationship preferred approach prediction not clear. To address this problem, authors aimed implement compare existing interpretation approaches, local interpretable agnostic explanation (LIME), shapely additive (SHAP) permutation feature importance by employing extreme boosting (XGBoost). Experiment conducted on dataset aim investigating most influencing output. Overall, experimental evidently appears reveal that blood glucose has impact outcome.