Leveraging Shapley Additive Explanations for Feature Selection in Ensemble Models for Diabetes Prediction

2024

Using SHAP for Better Diabetes Prediction

Sample size: 768 publication 10 minutes Evidence: high

Author Information

Author(s): Prasant Kumar Mohanty, Sharmila Anand John, Rabindra Kumar Barik, Diptendu Sinha Roy, Manob Jyoti Saikia

Primary Institution: Biomedical Sensors & Systems Lab, University of Memphis, Memphis, TN, USA

Can Shapley Additive explanations improve feature selection in machine learning models for diabetes prediction?

The study found that using the top three influential features significantly enhances the performance of diabetes prediction models.

Using SHAP values, the study identified the top three features that significantly influence diabetes predictions.
Ensemble models using the top three features achieved an AUC of 0.94.
Performance metrics improved when focusing on the most influential features compared to using all features.

This study shows that picking the most important features can help computers predict diabetes better and faster.

The study used ten different machine learning algorithms to evaluate diabetes prediction performance based on various feature sets.

Potential biases may arise from the dataset's demographic limitations.

The study's findings may not generalize to all populations due to the specific dataset used.

The dataset includes female patients from the Pima Indian community, aged 21 years or older.

0.001

95%

p<0.05

Access the complete publication on the publisher's website