[Paper Review] Dataset Optimization for Chronic Disease Prediction with Bio-Inspired Feature Selection

Abeer Dyoub, Ivan Letteri|arXiv (Cornell University)|Dec 17, 2023

Artificial Intelligence in Healthcare5 citations

TL;DR

This study proposes a bio-inspired feature selection framework using Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Whale Optimization Algorithm (WOA) to reduce dimensionality in high-dimensional chronic disease datasets, improving model accuracy and interpretability. GA outperformed other algorithms, achieving a 2.8% accuracy gain in breast cancer prediction while reducing features from 30 to 12 and cutting training time by 55%.

ABSTRACT

In this study, we investigated the application of bio-inspired optimization algorithms, including Genetic Algorithm, Particle Swarm Optimization, and Whale Optimization Algorithm, for feature selection in chronic disease prediction. The primary goal was to enhance the predictive accuracy of models streamline data dimensionality, and make predictions more interpretable and actionable. The research encompassed a comparative analysis of the three bio-inspired feature selection approaches across diverse chronic diseases, including diabetes, cancer, kidney, and cardiovascular diseases. Performance metrics such as accuracy, precision, recall, and f1 score are used to assess the effectiveness of the algorithms in reducing the number of features needed for accurate classification. The results in general demonstrate that the bio-inspired optimization algorithms are effective in reducing the number of features required for accurate classification. However, there have been variations in the performance of the algorithms on different datasets. The study highlights the importance of data pre-processing and cleaning in ensuring the reliability and effectiveness of the analysis. This study contributes to the advancement of predictive analytics in the realm of chronic diseases. The potential impact of this work extends to early intervention, precision medicine, and improved patient outcomes, providing new avenues for the delivery of healthcare services tailored to individual needs. The findings underscore the potential benefits of using bio-inspired optimization algorithms for feature selection in chronic disease prediction, offering valuable insights for improving healthcare outcomes.

Motivation & Objective

• To enhance predictive accuracy and model interpretability in chronic disease classification by reducing high-dimensional medical data.
• To evaluate the effectiveness of bio-inspired optimization algorithms—GA, PSO, and WOA—for feature selection in diverse chronic disease datasets.
• To investigate how feature selection impacts model performance across metrics like accuracy, precision, recall, and F1 score.
• To identify optimal algorithmic configurations and fitness function designs that balance feature reduction and predictive performance.
• To support early intervention and precision medicine by enabling more efficient, reliable, and actionable disease prediction models.

Proposed method

• Employed three bio-inspired optimization algorithms—Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Whale Optimization Algorithm (WOA)—for feature selection in chronic disease datasets.
• Designed a fitness function that prioritizes classification accuracy, with the goal of identifying the minimal subset of features that maintains or improves model performance.
• Applied the selected feature subsets to multiple classifiers (e.g., Decision Tree, SVM, Logistic Regression, Neural Network) to evaluate generalization across model types.
• Used a neural network with two hidden layers (10 neurons each) to assess training efficiency and convergence behavior under reduced feature sets.
• Conducted comparative analysis across four chronic diseases—diabetes, cancer, kidney, and cardiovascular diseases—using standardized performance metrics.
• Performed extensive data pre-processing to ensure data quality and reliability before feature selection and model training.

Experimental results

Research questions

• RQ1 How do GA, PSO, and WOA compare in reducing feature count while maintaining or improving classification accuracy for chronic disease prediction?
• RQ2 What is the impact of feature selection on model training time and convergence behavior, particularly in neural networks?
• RQ3 To what extent does the fitness function’s focus on accuracy alone affect performance on other metrics like precision, recall, and F1 score?
• RQ4 How does dataset-specific characteristics (e.g., number of features, class distribution) influence the effectiveness of each bio-inspired algorithm?
• RQ5 Can feature selection via bio-inspired methods lead to clinically interpretable and actionable predictions in chronic disease management?

Key findings

• Genetic Algorithm (GA) achieved the highest overall performance, outperforming WOA and PSO in 50% of cases for F1 score and demonstrating the best balance of accuracy and feature reduction.
• In the breast cancer dataset, feature reduction from 30 to 12 features increased accuracy by 2.8% and reduced training time by 55%, highlighting significant efficiency gains.
• PSO struggled to improve fitness across generations, likely due to suboptimal initialization of parameters (w, c1, c2), indicating sensitivity to hyperparameter tuning.
• Despite focusing only on accuracy in the fitness function, GA maintained strong performance across all metrics, with F1 scores exceeding the baseline in 50% of cases.
• Training time did not always decrease with fewer features—especially in the heart failure dataset, where reduced features led to longer training cycles, suggesting model architecture sensitivity.
• WOA showed the most consistent performance in reducing feature count (down to 5 features), but its accuracy and F1 scores were lower than GA and PSO in most cases.