A Machine Learning Classification Technique for Predicting Prostate Cancer

This paper presents and validates various classification techniques on supervised machine learning (ML) for predicting prostate cancer. A modified Logistic Regression (LR) classifier is proposed and implemented on patients who are susceptible to prostate cancer. The proposed classification technique uses both clinical and tumor stage characteristics. Clinical characteristics considered are BMI, age, cystitis infections, and smoking history. Tumor stage characteristics are stages of Tumor Node Metastasis (TNM), American Joint Committee on Cancer (AJCC) and Prostate Specific Antigen (PCA). Results obtained show improvement in accuracy and positive prediction value (PPV) as compared to existing classifiers. Results are compared and validated with performance measures of Specificity (Sp) and Sensitivity (Se), recording a minimum of 3% improvement in Pc prediction accuracy. The implemented ML classification technique also shows a clinical impact on Pc diagnosis with a 4 % improvement in Sp.