DEVELOPMENT OF AN EXPLAINABLE HYBRID MACHINE LEARNING FRAMEWORK FOR PROSTATE CANCER PREDICTION

AMECHI, ELOCHUKWU WESLEY (2026) DEVELOPMENT OF AN EXPLAINABLE HYBRID MACHINE LEARNING FRAMEWORK FOR PROSTATE CANCER PREDICTION. Other thesis, Godfrey Okoye University, Enugu Nigeria.

Text Amechi Elochukwu Wesley Project.PDF Download (2MB)

Abstract

Prostate cancer accounts for the most male deaths due to cancer in Nigeria; however, the country suffers from lack of availability of accurate diagnosis infrastructure, especially in non-urban settings. Available AI-supported prostate cancer prediction systems rely heavily on imaging techniques such as Magnetic Resonance Imaging (MRI) and digital pathology slides, which make them hard to implement at locations lacking adequate infrastructure. The current research focuses on filling this void by designing a web-based machine learning system, ProstaPredict AI, for predicting the result of prostate cancer biopsies using clinical structured tabular data. This project used the Agile development approach and leveraged the Flask framework with an XGBoost classification model that was trained on 23 clinical and demographic variables. Data processing and scaling were done by the combination of StandardScaler and OneHotEncoder from scikit-learn package, while SHAP TreeExplainer was incorporated to provide visualization of important features along with each prediction. Data leakage test was performed to check the correctness of the feature selection process, and three models were tested using five performance criteria prior to deploying the XGBoost classifier. These three models all achieved random classification performance when evaluated with proper care in a leak-free environment, as indicated by their Reciever Operating Characteristics-Area Under Curve (ROC-AUC) values of around 0.50. This was believed to be because of the synthetic data used in training, as opposed to a failure on part of the system design. From the leakage test, it can clearly be seen that post-diagnosis attributes boosted the AUC of our models to above 0.97. ProstaPredict AI showcases how an explainable AI decision support system that is lightweight, role-based, and explainable can be developed and implemented without any specialized infrastructure setup. This system gives a proven platform that is instantly applicable on clinical data and develops an approach to identify data leakage problems in clinical datasets. It needs to be developed further for retraining on annotated biopsy data from Nigerian health facilities to improve prediction performance.

Item Type:	Thesis (Other)
Subjects:	Q Science > Q Science (General)
Divisions:	Faculty of Computing And Information Technology (FACIT)
Depositing User:	Cynthia Ugwuoti
Date Deposited:	24 Jul 2026 12:19
Last Modified:	24 Jul 2026 12:19
URI:	http://eprints.gouni.edu.ng/id/eprint/5955

Actions (login required)

View Item