Personalized Healthcare: Utilizing Probabilistic Models for Disease-Drug Association and Treatment Prediction

The current healthcare system faces challenges in delivering treatment recommendations personalized to individual patient needs, leading to issues such as misdiagnosis, delayed treatment plans, and harmful drug interactions. In response to these challenges, personalized medicine has emerged as a critical component in healthcare, aiming to provide patient-specific clinical treatment recommendations based on individual conditions, thereby enabling more accurate diagnoses and treatments. Machine Learning (ML) has proven to be a crucial approach in advancing personalized medicine and healthcare. Numerous ML algorithms have been implemented to generate suitable recommendations tailored to individual patient conditions. However, most of these algorithms lack explain ability and reasoning behind their decisions, often relying on advanced black-box models. In this paper, we implement the Bayesian networks algorithm, which utilizes a probabilistic learning approach that is both explainable and effective due to its ability to learn, represent relationships, and exploit correlations between variables, thereby enabling ethically informed predictions of risks and side effects. Using graphical models, healthcare providers can deliver individualized care by proposing methods and treatment plans adapted to each patient's specific needs and conditions. This approach aims to minimize side effects and provide precise treatment recommendations, thereby enhancing overall patient care. We conducted three experiments to develop explainable predictive models, exploring three predictive classes: drug class, drug activity, and the side effects of drugs associated with diseases such as Allergies, Alzheimer's, Cancer, and Stroke, etc. The predictive models achieved high accuracy rates, ranging from 82% to 99%, and obtained very reasonable validation accuracies.