PREDICTING PROSTATE CANCER WITH MACHINE LEARNING: FINDING A BALANCE BETWEEN ACCURACY AND COMPUTATIONAL EFFICIENCY

Abstract

Background: One of the most prevalent cancers in men is prostate cancer, and early identification is crucial to reducing death and enhancing treatment results. Traditional diagnostic methods could be laborious and imprecise. Machine learning (ML) techniques can reduce computer overhead and improve and automate prostate cancer prediction.

Methods: In this study, clinical criteria such as area, smoothness, texture, radius, and perimeter were incorporated in the improved prostate cancer dataset. Preprocessing techniques included label encoding, feature scaling, and mutual information to choose the top ten informative features. Several machine learning models, such as Extreme Gradient Boosting (XGBoost), Random Forest, K-Nearest Neighbours (KNN), Dark Gradient Boosting Machine (LightGBM), and Logistic Regression, were trained and assessed. To adjust the hyperparameters, the Optuna optimisation framework was employed. Computation time (training and testing) and accuracy were used to evaluate performance.

Results: Each model had outstanding forecasting ability. LightGBM fared better than KNN (98.33%) and XGBoost (99.16%) with an accuracy of 99.37%. KNN was the most efficient model with a total runtime of 0.038 seconds; LightGBM and XGBoost were somewhat slower but provided higher accuracy. Even with a moderate accuracy rate of 87.22%, logistic regression remained computationally efficient.

Conclusion: The findings demonstrate that KNN provides the best balance between accuracy and time efficiency, making it suitable for rapid diagnostic applications. However, LightGBM and XGBoost remain superior choices when accuracy is essential. The proposed approach demonstrates how well-suited optimised machine learning models may be as predictive tools for prostate cancer by encouraging early diagnosis and aiding therapeutic decision-making.