What Effect Do Interpretable Machine Learning Models Have on Accuracy and Transparency in Healthcare?
Online Web Editor: Haamid Bala
Author: Sheikh Anas Neyaz
Date: 13 August 2025
⸻
Abstract
This study explores how explainable AI (XAI) models can enhance transparency in healthcare while maintaining high accuracy. Using 356 anonymized records from five regional hospitals, datasets included BMI, cholesterol, glucose levels, and blood pressure.
Three machine learning models — XGBoost, Random Forest, and Deep Neural Network (DNN) — were trained to predict diabetes and cardiovascular diseases.
Interpretability was achieved using SHAP (SHapley Additive exPlanations) values, which assigned numerical weightage to factors influencing predictions.
The XGBoost model achieved 93.1% accuracy with precision and recall of 0.92 and 0.90, respectively. The DNN model, while attaining 95.2% accuracy, lacked interpretability.
By integrating SHAP, transparency ratings improved by 41% among 26 professionals and 44% among 28 patients. The top predictive contributors were glucose (23.4%), BMI (18.7%), and systolic pressure (16.5%).
Benchmark analysis showed that interpretable models retained 96% of DNN’s accuracy while significantly outperforming it in transparency and trust evaluations.
⸻
Introduction
Artificial Intelligence (AI) has evolved from a futuristic concept to a transformative force across industries — particularly healthcare. From aiding surgeries to predicting diseases, AI now plays a critical role in clinical decision-making.
However, despite its computational power, AI adoption in medicine faces a major hurdle — a lack of transparency and interpretability. Healthcare professionals are hesitant to rely on systems that cannot clearly explain their conclusions.
This study focuses on interpretable machine learning models, designed not only for accuracy but also for explainability. The research involved 356 anonymized records from hospitals in Ganderbal, Sopore, Beehama, and Srinagar, examining how explainable AI can balance performance with trustworthiness.
⸻
Methodology
Patient data included key health indicators: BMI, cholesterol, glucose levels, and blood pressure (systolic and diastolic).
Three predictive models were trained:
• XGBoost
• Random Forest
• Deep Neural Network (DNN)
The goal was to forecast the likelihood of diabetes and cardiovascular diseases — conditions where early intervention can save lives.
Interpretability was incorporated using SHAP (SHapley Additive exPlanations), applied to XGBoost and Random Forest models. SHAP quantified each feature’s influence on predictions, offering transparency for medical decision-making.
A survey involving 26 healthcare professionals and 28 patients assessed trust levels based on accuracy, clarity, and perceived assurance.
⸻
Results
• DNN Model: 95.2% accuracy (highest overall)
• XGBoost Model: 93.1% accuracy
• Random Forest Model: 92.3% accuracy
SHAP-integrated models achieved:
• Precision: 0.91 (XGBoost), 0.89 (Random Forest)
• Recall: 0.90 (both models)
Transparency and trust assessments revealed:
• 41% increase in professional confidence
• 44% increase in patient trust (n=54)
Top predictive contributors:
1. Glucose — 23.4%
2. BMI — 18.7%
3. Systolic Pressure — 16.5%
Interpretable models maintained 96% of DNN’s accuracy while significantly improving clarity, ethical accountability, and user trust.
⸻
Conclusion
As AI becomes deeply embedded in healthcare, interpretability is no longer optional — it is essential. Transparent models bridge the gap between machine precision and human empathy, reinforcing ethics, accountability, and trust in medical technology.
Interpretable AI ensures that technology remains not just powerful, but also responsible and humane — reflecting the true spirit of healthcare.
⸻
📧 For correspondence: neyazanas08@gmail.com
🔖 Tags: #HealthcareAI #MachineLearning #ExplainableAI #Transparency #KashmirPen
