More and more people are caring about the health and medical diagnosis problems. However according to the administrations report more than 200 thousand people in China even 100 thousand in USA die each year due to medication errors. More than 42% medication errors are caused by doctors because experts write the prescription according to their experiences which are quite limited. Technologies as data mining and recommender technologies provide possibilities to explore potential knowledge from diagnosis history records and help doctors to prescribe medication correctly to decrease medication error effectively. In this paper we design and implement a universal medicine recommender system framework that applies data mining technologies to the recommendation system. The medicine recommender system consists of database system module data preparation module recommendation model module model evaluation and data visualization module. We investigate the medicine recommendation algorithms of the SVM (Support Vector Machine) BP neural network algorithm and ID3 decision tree algorithm based on the diagnosis data. Experiments are done to tune the parameters for each algorithm to get better performance. Finally in the given open dataset SVM recommendation model is selected for the medicine recommendation module to obtain a good trade-off among model accuracy model efficiency and model scalability. We also propose a mistake-check mechanism to ensure the diagnosis accuracy and service quality. Experimental results show our system can give medication recommendation with an excellent efficiency accuracy and scalability.



SVM-naïve bayes-random forest-DNN-Multinomial Logistic Regression


machine learning


₹12000 (INR)


IEEE 2016