Creation of Anomaly Detection Model for IoT Thermometer Used in Hospital Refrigerators

Abstract

Hospital environments need hospital refrigerators to store drugs, vaccines, blood bags, among others. Such equipment is configured in order to maintain a certain temperature range, since the stored products are sensitive to temperature changes beyond this range. This project aims to analyse the temperature variations beyond adequate. In the experiments performed, different anomaly detection techniques were implemented, using three clustering methods: k-means, DBSCAN and Isolation Forest. Taking into account the accuracy found (76.7%), the method used was DBSCAN. With the analysis performed, it was possible to see several relationships between the temperature values, the number of alerts and the times they happened. It was observed that most of the anomalies found happened between 6:00 and 8:00 am, coinciding with the shift change time between employees.