COMPARATIVE PERFORMANCE EVALUATION OF K-NEAREST NEIGHBOR (KNN) AND LEARNING VECTOR QUANTIZATION (LVQ) FOR BREAST CANCER SURVIVAL PREDICTION USING THE HABERMAN CANCER SURVIVAL DATASET
DOI:
https://doi.org/10.5281/Keywords:
K-Nearest Neighbor; Learning Vector Quantization; Breast Cancer Survival Prediction; Haberman Survival Dataset; Machine Learning; Medical Classification; Healthcare AnalyticsAbstract
Background: Machine learning has become an essential tool for developing intelligent clinical
decision-support systems that assist healthcare professionals in disease diagnosis and prognosis.
Among supervised classification algorithms, K-Nearest Neighbor (KNN) and Learning Vector
Quantization (LVQ) are widely used because of their simplicity, interpretability, and suitability for
structured biomedical datasets. However, comparative evaluations of these algorithms under identical
preprocessing, optimization, and validation conditions remain limited, particularly for breast cancer
survival prediction. Objective: This study comparatively evaluated the predictive performance of K
Nearest Neighbor (KNN) and Learning Vector Quantization (LVQ) for breast cancer survival
prediction using the Haberman Survival Dataset from the UCI Machine Learning Repository.
Methods: The study utilized the Haberman Survival Dataset containing 306 patient records with three
predictor variables (patient age, year of operation, and number of positive axillary lymph nodes) and a
binary survival outcome. Data preprocessing involved data verification, Min-Max normalization, and
stratified data partitioning. Hyperparameter optimization was performed by testing multiple values of
K for KNN, while LVQ parameters, including learning rate, prototype initialization, and training
iterations, were optimized. Model performance was evaluated using Accuracy, Precision, Recall,
Specificity, F1-score, Matthews Correlation Coefficient (MCC), Receiver Operating Characteristic
Area Under the Curve (ROC-AUC), confusion matrices, computational execution time, and
McNemar's statistical test. Results: Both algorithms successfully classified breast cancer survival
outcomes. However, KNN consistently outperformed LVQ, achieving higher Accuracy, Precision,
Recall, F1-score, MCC, and ROC-AUC, indicating superior predictive performance. Conversely, LVQ
demonstrated faster prediction time and lower computational overhead due to its prototype-based
learning mechanism. McNemar's test further confirmed the superior classification performance of
KNN under the adopted experimental conditions. Conclusion: KNN demonstrated better predictive
capability for breast cancer survival prediction, whereas LVQ provided greater computational
efficiency. These findings suggest that KNN is more appropriate for accuracy-driven clinical decision support systems, while LVQ may be preferable in resource-constrained environments. Future research
should validate these findings using larger multicenter datasets, optimized LVQ variants, ensemble
learning methods, and explainable artificial intelligence techniques.
