Understanding Prostate Cancer Risk Using Statistical and Machine Learning Approaches: A Comparative Methodological Analysis

OpenAlex English

Karşılaştırmalı Metodolojik AnalizProstate cancer remains one of the most common and deadly malignancies among men worldwide, necessitating accurate risk prediction tools to enhance early diagnosis and personalized care.This study aims to compare the predictive capacity of traditional binary logistic regression with that of contemporary machine learning (ML) algorithms: support vector machines (SVM), K-nearest neighbors (KNN), chi-squared automatic interaction detection (CHAID), and C5.0 in identifying key risk factors and classifying prostate cancer status.A total of 501 male participants (248 diagnosed cases, 253 controls) were evaluated using a structured, 20-item questionnaire capturing demographic, clinical, and lifestyle parameters.Across all models, variables such as age, smoking status, and family history of cancer consistently emerged as significant predictors.Additional risk indicators included blood in semen or urine, frequency of urination, and daily activity levels.The classification accuracy achieved by each model was as follows: logistic regression (92.2%),SVM (89.92%),KNN (88.48%),CHAID (91.36%), and C5.0 (88%).Receiver operating characteristic analysis and cumulative gain curves confirmed the superior performance of logistic regression, achieving the highest accuracy (92.2%) and estimated area under the curve (92.2%) based on confusion matrix metrics.While logistic regression demonstrated optimal performance and interpretability for structured clinical data, ML models offered complementary insights by uncovering complex, nonlinear associations.The integration of statistical and ML methodologies may thus enhance clinical decision-making and contribute to the development of robust, data-driven diagnostic frameworks in prostate cancer care.