SGA-DT: An adaptive fusion framework for missing data imputation and interpretable healthcare classification

Area of Science:

• Machine Learning
• Healthcare Analytics
• Data Science

Background:

• Missing data is a significant challenge in healthcare analytics, potentially compromising model accuracy and clinical reliability.
• Improper imputation of missing values in sensitive healthcare data can lead to biased outcomes and delayed interventions.

Purpose of the Study:

• To propose SGA-DT, an adaptive and interpretable learning framework for robust healthcare prediction.
• To address the critical challenge of handling missing values in healthcare datasets.

Main Methods:

• SGA-DT combines genetically optimized support vector regression (SVR) with a decision tree (DT) classifier.
• It adaptively selects imputation strategies (SVR, iterative SVR, KNN+SVR) based on missingness levels.
• A genetic algorithm (GA) optimizes SVR kernel selection and hyperparameters.

Main Results:

• SGA-DT consistently outperforms other integrated frameworks in accuracy, precision, recall, and F-measure across multiple datasets.
• The framework demonstrates robustness and generalizability in healthcare prediction tasks.
• Interpretability analysis using decision trees supports clinical transparency under varying missingness levels.

Conclusions:

• SGA-DT offers a robust, interpretable, and generalizable solution for healthcare prediction with missing data.
• The adaptive imputation strategy and genetic optimization enhance prediction performance.
• The framework contributes to more reliable and transparent clinical decision-making.