Cure-Sne Model Detects Outliers in Data to Optimize the Process of Clustering Using Representatives

Stunting is a problem that always appears in our midst. Stunting prevalence data is important for observing and monitoring the condition of stunting that occurs. However, data with bad values affects the accuracy of the data, so it has a big impact on the information that will bereceived when the data is processed. For this reason, stunting data is processed using the Cure-Sne model to group stunting prevalence which produces good information. With the Cure- Sne model, the processed dataset also detects a number of outiers that interfere with the values of other real data. In the Cure-Sne process, the amount of data used is 120K rows with5 attributes which are processed with Cure-Sne to produce 4 Clusters. In Cluster 1, the predominant gender is male, and the nutritional status is classified as normal. Cluster 2 shows a predominance of females with a severely undernourished status. In Cluster 3, males dominate with a normal nutritional status, while Cluster 4 consists mainly of females, also with a normal nutritional status. From 120,000 rows in the dataset, many data were detected as outliers. There are 6,060 outlier data points identified from a total of 120,000 records with 4 attributes. This indicates that a significant portion of the datamay not conform to the expected patterns or distributions, suggesting the presence of anomalies that warrant further investigation. Evaluation of Cure-Sne is carried out with the Sillhoute Score. The evaluation or accuracy of the Cure-Sne model in analyzing stunting data is determined to be 0.8450, equivalent to 84.50%.