Asian Journal of Probability and Statistics

Terminal diseases, including kidney disease, heart failure, and cancer, represent a formidable global health challenge, claiming millions of lives annually and placing immense strain on healthcare systems worldwide. In resource-limited settings such as Nigeria, where access to advanced diagnostic technologies and specialized care is often constrained, these conditions are particularly devastating, contributing to high morbidity and mortality rates due to delayed detection and intervention. This study addresses this critical gap by developing and rigorously evaluating a machine learning model designed to predict the risk of terminal diseases, leveraging the predictive power of Artificial Neural Networks (ANN) and Random Forest algorithms. The research utilizes a clinical dataset sourced from Ondo State Teaching Hospital, comprising 1200 instances from 100 patients, with 750 cases of terminal diseases and 450 non-terminal cases, characterized by 24 input attributes such as age, blood pressure, and hemoglobin levels.

The ANN model, constructed with a 25-neuron input layer, a 56-node hidden layer, and a single output neuron, achieved a testing accuracy of 98%, outperforming the Random Forest model, which recorded accuracies of 85.25% for heart disease, 99% for kidney disease, and 98.25% for cancer. These results highlight the ANN’s superior ability to generalize across multiple diseases, supported by additional metrics such as a 98.7% recall and a 99.2% AUC-ROC, affirming its precision and discriminative capacity. The Random Forest, while highly effective for kidney disease, showed variability across conditions, suggesting differential strengths in handling disease-specific patterns. Both models were developed using a robust methodology, including data preprocessing with Min-Max scaling and LabelEncoder, an 80/20 train-test split, and k-fold cross-validation to ensure reliability.

This high accuracy demonstrates the potential of the ANN-based model as a decision-support tool for early diagnosis, offering a scalable, cost-effective solution to enhance healthcare delivery in settings where traditional diagnostics are limited. By enabling clinicians to identify at-risk patients sooner, the model could facilitate timely interventions, potentially reducing the progression of terminal diseases and improving patient outcomes. This paper provides a detailed discussion of the methodology, including dataset preparation, model architecture, and training processes, alongside a comprehensive analysis of performance metrics such as accuracy, precision, and F1-score. It also explores the implications for healthcare delivery, emphasizing the model’s relevance in resource-constrained environments and its capacity to address Nigeria’s pressing health challenges.