Addressing Binary Classification over Class Imbalanced Clinical Datasets Using Computationally Intelligent Techniques

Area of Science:

• Machine Learning in Healthcare
• Data Science
• Clinical Informatics

Background:

• Clinical datasets are crucial for intelligent healthcare systems.
• Real-world clinical data often exhibits class imbalance, leading to poor classifier performance.
• This imbalance negatively impacts accuracy, precision, and recall, necessitating effective data balancing techniques.

Purpose of the Study:

• To empirically evaluate the performance of six classifiers on five imbalanced clinical datasets.
• To compare seven distinct class balancing techniques for addressing data imbalance in clinical machine learning.
• To identify optimal strategies for handling imbalanced clinical datasets in supervised learning.

Main Methods:

• Literature review on class imbalanced learning.
• Performance evaluation of Decision Tree, k-Nearest Neighbor, Logistic Regression, Artificial Neural Network, Support Vector Machine, and Gaussian Naïve Bayes classifiers.
• Application and comparison of seven balancing techniques: Undersampling, Random Oversampling, SMOTE, ADASYN, SVM-SMOTE, SMOTEEN, and SMOTETOMEK.

Main Results:

• SMOTEEN consistently outperformed the other six data-balancing techniques across all tested classifiers and datasets.
• The remaining six balancing techniques showed comparable, yet moderately lesser, performance compared to SMOTEEN.
• Analysis explored the reasons behind the effectiveness of specific classifiers and balancing methods.

Conclusions:

• SMOTEEN is a highly effective technique for addressing class imbalance in clinical datasets.
• The findings provide practical recommendations for improving supervised machine learning model performance in healthcare.
• Careful selection of data-balancing techniques is essential for robust clinical predictive models.