Diabetes Detection Models in Mexican Patients by Combining Machine Learning Algorithms and Feature Selection...

Area of Science:

• Medical Informatics
• Computational Biology
• Health Data Science

Background:

• Diabetes mellitus is a significant global health concern requiring accurate diagnostic tools.
• Machine learning (ML) models are increasingly used for diabetes detection, but performance depends on data quality and algorithm choice.
• Feature selection is crucial for optimizing ML models by identifying the most relevant input data for accurate classification.

Purpose of the Study:

• To investigate the integration of feature selection techniques with ML classifiers for improved diabetes detection.
• To evaluate the performance of models using Akaike information criterion and genetic algorithms for feature selection.
• To compare the efficacy of these optimized models against existing approaches in diabetes diagnosis.

Main Methods:

• Employed two feature selection techniques: Akaike information criterion and genetic algorithms.
• Integrated selected features with six ML classifier algorithms: support vector machine, random forest, k-nearest neighbor, gradient boosting, extra trees, and naive Bayes.
• Utilized clinical and paraclinical features from diverse datasets for model training and evaluation.

Main Results:

• Achieved superior performance with diagnostic accuracies surpassing 94%.
• Demonstrated that feature selection enables effective model development using reduced datasets.
• Validated the significant role of feature selection in enhancing the predictive power of diabetes detection models.

Conclusions:

• Feature selection is pivotal for improving the accuracy and efficiency of ML-based diabetes detection models.
• The proposed approach advances medical diagnostic capabilities, aiding healthcare professionals in diabetes diagnosis and treatment decisions.
• Optimized models offer a more robust and data-efficient solution for supporting clinical decision-making in diabetes management.