Explainable Machine Learning for Efficient Diabetes Prediction Using Hyperparameter Tuning, SHAP Analysis, Partial Dependency, and LIME
Open Access
- 12 December 2024
- journal article
- research article
- Published by Wiley in Engineering Reports
- Vol. 7 (1)
Abstract
Diabetes is a chronic metabolic disease characterized by elevated blood glucose levels and poses significant health risks, such as cardiovascular disease and cognitive damage. Understanding the causes of diabetes is crucial to managing it and preventing complications. The clinical community has a lot of diabetes diagnostic data. Machine learning algorithms may simplify finding hidden patterns, retrieving data from databases, and predicting outcomes. To tackle the challenge of designing an improved diabetes classification algorithm that is more accurate, random oversampling and hyper‐tuning parameter techniques have been used in this study. Whereas most of the existing methods were built upon considering any single dataset, for getting more acceptability in general, our proposed model has been designed based on two benchmark datasets: the BRFSS dataset, which has multiple classes, and the Diabetes 2019 dataset, which has binary classes. What is more, to improve the comprehensibility of the proposed model, a variety of explainability methodologies such as SHapley Additive Explanations (SHAP), Partial Dependency, and Local Interpretable Model‐agnostic Explanations (LIME) have been implemented which are not often noticed in the previous works. The detailed explainability charts will enable the end users or practitioners to understand the exact factors of any given diagnostic report. This research focused on classifying type 2 diabetes using machine learning and providing an explanation for the outcomes derived from the model predictions. Random oversampling and quantile transform are used to rectify imbalances in the dataset and guarantee the resilience of model training. By meticulously adjusting parameters with gridsearchCV, we successfully optimized our models to attain exceptional accuracy across binary and multi‐class datasets. We evaluate the proposed model using two datasets and performance metrics. The extra trees classifier (ET) performed exceptionally, achieving 97.23% accuracy on the multi‐class dataset and 97.45% on the binary dataset.Keywords
Funding Information
- Jagannath University (JnU/Gobeshona/GaPro/2022–23/Science/18)
This publication has 43 references indexed in Scilit:
- Relating the Partial Dependence Plot and Permutation Feature Importance to the Data Generating Process2023
- Machine Learning Based Diabetes Detection Model for False Negative ReductionBiomedical Materials & Devices, 2023
- A novel method for diabetes classification and prediction with PycaretMicrosystem Technologies, 2023
- An assessment of machine learning models and algorithms for early prediction and diagnosis of diabetes using health indicatorsHealthcare Analytics, 2022
- Machine learning-based lung and colon cancer detection using deep feature extraction and ensemble learningExpert Systems with Applications, 2022
- Extra-Tree Classifier with Metaheuristics Approach for Email Classification2019
- Diabetes detection using deep learning algorithmsICT Express, 2018
- A Brief Survey on Random Forest Ensembles in Classification Model2018
- Design of experiments and response surface methodology to tune machine learning hyperparameters, with a random forest case-studyExpert Systems with Applications, 2018
- KNN Model-Based Approach in Classification2003