Sofiene Haboubi, Abir Ben Cheikh
Patient similarity analysis is a prerequisite for applying machine learning technology to medical data. The subject of this article is “Similarity of patients in predictive models using medical data”. This article describes, how to create a system capable of analyzing patient data. We first implemented a methodology to extract useful information from raw data. We then determined, for the set of data extracted from a real database from a medical practice, the similarities that may exist between patients; based on several explanatory variables. We then derived a meaningful distance metric to measure the similarity between the patients represented by their key indicators. Thus, we have proven the importance of defining strong associations between determined attributes. We have developed a process to select the best attributes that will lead to the prediction. Finally, and after grouping the patients using the partition clustering approach (the k-medoid algorithm), we built a predictive linear regression model. For the learning phase, we combined different supervised and unsupervised techniques. We have chosen medical prescription as the area of application to predict the right medication for the patient. The results obtained show that the proposed approach can produce good predictions.
Data Mining, Machine Learning, Medical Informatics, Similarity Measures, Multiple Linear Regression Model, k-Medoids Clustering
Cite this paper
Sofiene Haboubi, Abir Ben Cheikh. (2021) Similarity of Patients in Predictive Models Using Medical Data: Case of Auto-prescription Drugs for Diabetic Patients. International Journal of Computers, 6, 33-39