Combination of Local Outlier Factor and Winsorization for Clustering Outlier in Medical Records

In medical records, the doctors need to stay updated about the condition of their patients, but it would make a problem for the medical records. The medical records would be inconsistent due to changes in information regarding the result of examinations and leads to data that contains noises and outliers. But in medical records, the data sometimes can be mistaken as noise and it cannot be deleted as it is, because the medical records might contain one or more important information. In this research, the medical records are taken from one of the Regional Public Hospital in North Sumatera with limited work of ethics, and efforts will be made to solve the outlier problem in medical records by clustering the outliers into normal cluster by combining the Local Outlier Factor and Winsorization technique. The dataset consists 2 medical record data which contains data on outpatients and inpatients with various types of diagnoses in the last 3 years, and the variable that will be used are Gender, Date of Birth, and the Diagnosis, but it will be reduced into 2 variables, which is Diagnosis and Age (Years Old). In the end, the results are evaluated using a Confusion Matrix to see the percentage of performance for the proposed combination between the Local Outlier Factor method and the Winsorization technique, and the results are: Precision 96.8%, Recall 99.8%, Accuracy 96.9%, F1 Score 98.3%. The results have achieved the best percentage between the K-values.