Evaluating PCA and LDA to Improve Machine Learning Classification of Consumer Behavior in Health Informatics

Abstract

Behavioral data derived from consumer activity offer significant potential in health informatics, particularly for developing predictive models related to lifestyle, adherence, and patient engagement. This study evaluates the effect of two widely used dimensionality reduction techniques, namely principal component analysis (PCA) and linear discriminant analysis (LDA) on the performance of five supervised machine learning classifiers: logistic regression (LR), support vector machine (SVM), naive bayes (NB), decision tree (DT), and random forest (RF). The experimental dataset, although sourced from a commercial context, contains demographic and economic attributes commonly found in health-related behavioral data, such as age and income. Results indicate that LDA significantly improves classification performance across all models, with Random Forest achieving the highest scores: accuracy = 0.91, precision = 0.88, recall = 0.85, F1-score = 0.86, and AUC = 0.95 when trained on LDA-transformed features. SVM also performed competitively under the same configuration (AUC = 0.94). Conversely, PCA provided moderate gains but underperformed in capturing class-discriminative information compared to LDA. These findings demonstrate that integrating LDA with robust classifiers such as RF and SVM enhances both predictive accuracy and model interpretability, offering practical benefits for behavior-informed health decision support systems. The study highlights the relevance of supervised feature transformation in optimizing data pipelines for personalized healthcare applications.