Comparing Linear and Quadratic Discriminant Analysis for Cervical Cancer Risk Prediction Using Behavioral Attributes

Cervical cancer remains a critical global health challenge, yet early prediction using non-medical determinants is underexplored. This study aims to compare linear and quadratic discriminant analysis for early cervical cancer risk assessment based on behavioral and psychosocial factors. A quantitative experimental approach was utilized, analyzing a publicly available behavioral dataset of seventy-two instances and eighteen psychosocial features. Following rigorous statistical validation, feature normalization, and stratified data splitting, both discriminant models were trained and evaluated using accuracy and area under the receiver operating characteristic curve metrics. The empirical findings reveal a striking performance disparity between the classifiers. Linear discriminant analysis emerged as the superior model, achieving an exceptional accuracy of 93.33% and an area under the curve of 0.9464, demonstrating robust discriminative capability. Crucially, it attained perfect sensitivity with zero false negatives, eliminating the most dangerous diagnostic errors. Conversely, quadratic discriminant analysis exhibited complete predictive failure, performing no better than random guessing. This extreme divergence indicates that the underlying behavioral data possesses linearly separable characteristics. The additional complexity of estimating class-specific covariance matrices in the quadratic model resulted in severe overfitting and numerical instability given the limited sample size, proving that simpler, assumption-aligned models outperform complex alternatives. Ultimately, linear discriminant analysis provides an accurate, interpretable, and cost-effective decision support tool for identifying cervical cancer vulnerability. These findings validate that integrating social psychology with machine learning offers a robust framework for early disease screening in resource-constrained environments.