Medicine Recommendation System Based on Symptoms Using Machine Learning

The problem of access to prompt medical attention still represents a critical issue for many areas in the world, particularly for remote or low-income regions, in which availability of a doctor is exceptionally low. In a country such as India, for instance, where there exists one doctor per 834 people, patients are more prone to rely on self-medication with the ensuing potential for erroneous cures, side effects, or serious medical consequences. In order to remedy this issue, a Medicine Recommendation System was designed; based on patient symptoms, it suggests optimal medicines as well as details regarding dosage, side effects, and other useful information. Two datasets were utilized throughout this project: a symptom-prognosis dataset, containing 4920 training instances relating 41 disease classes and 132 binary symptom features, and a merged pharmaceutical database of 100,000 instances corresponding to 30 classes of drugs. The six machine learning algorithms (Random Forest, Gradient Boosting Classifier, Support Vector Machine, Decision Tree, K-Nearest Neighbour, Naive Bayes) trained were all tested against the training dataset and compared. Among all algorithms considered, Random Forest and Gradient Boosting demonstrated the highest accuracy of 97.62%; both algorithms proved highly effective in pinpointing the most indicative symptoms, identifying high fever, fatigue, and vomiting as the three most diagnostically relevant symptoms, and providing a set of detailed medical recommendations for each user case.