Binary classification with fuzzy logistic regression under class imbalance and complete separation in clinical...

Area of Science:

• Clinical studies
• Data science
• Biostatistics

Background:

• Binary classification in clinical studies faces challenges from imbalanced class distribution and complete separation.
• These issues lead to inaccurate predictions and biased results in patient classification.

Purpose of the Study:

• To introduce and evaluate a fuzzy logistic regression framework for binary classification in clinical settings.
• To address and mitigate the impact of class imbalance and complete separation on classification accuracy.

Main Methods:

• A fuzzy logistic regression framework utilizing triangular fuzzy numbers for coefficients, inputs, and outputs was developed.
• The framework produces crisp classification results, enhancing the handling of imbalanced and separated data.

Main Results:

• Fuzzy logistic regression demonstrated consistent high performance across twelve clinical datasets, with excellent sensitivity, specificity, F1, precision, and Mathew's correlation coefficient.
• The model showed no adverse impact from data imbalance or separation, outperforming classical logistic regression and ten other benchmark methodologies.

Conclusions:

• Fuzzy logistic regression offers a robust solution for binary classification in clinical studies, particularly when dealing with imbalanced and separated data.
• The framework provides accurate predictions and reliable insights for patient classification, enhancing clinical study outcomes.