Health Predictions Redefined: The Impact of AI on Future Disease Diagnosis

Abstract

Healthcare professionals often apply a one-size-fits-all approach in patient care, potentially leading to misdiagnosis, suboptimal treatments, and higher healthcare costs. Machine-learning models have garnered attention for their ability to improve diagnostic accuracy, with numerous studies focusing on machine learning applications for individual disease predictions, such as Type II diabetes, heart disease, kidney disease, and hypertension. However, limited research has tackled the combined prediction of Type I diabetes (standard cases), Type II diabetes (gestational diabetes), and cardiovascular disease, presenting a significant research gap.

To address this gap, we introduce a set of benchmark corpora based on authentic patient records, targeting specific disease categories. The first contribution is a heart disease corpus containing 606 instances. The second and third contributions consist of two separate corpora, each with 849 instances: one focused on standard diabetes cases and the other on gestational diabetes cases. We evaluated these corpora with ten machine-learning algorithms and five deep-learning algorithms, rigorously comparing their performance across common metrics, including accuracy, precision, recall, and $F_{1}$-score. Our results revealed high performance across all models, with top $F_{1}$-scores of 0.785 using Random Forest, 0.790 with Gradient Boosting, and 0.994 using BiLSTM for the combined disease prediction. These findings suggest that the proposed datasets and models provide a robust foundation for accurate and scalable high-risk disease prediction, contributing a valuable, multidimensional approach to personalized patient care