Evaluation of Jenks Natural Breaks Clustering Algorithm for Changepoint Identification in Streaming Sensor Data

This work evaluates the performance of a non-supervised clustering method for identifying abrupt changepoints in streaming sensor data. The proposed method utilizes the Jenks Natural Breaks (JNB) algorithm, applied in near real-time using sliding temporal windows to analyze sections of sensor data and identify instances of significant phase changes. It is suitable for sensing applications that rely on detecting instantaneous changes in the sensed data for fast decisions, such as fire alarms, fault detection, and activity recognition. The method was applied to a custom dataset from twelve electrodes transitioning between different materials. Performance was evaluated based on detection accuracy and delay comparisons. Results demonstrate that applying JNB in a sliding window with a step size of half its length achieves the highest detection accuracy and the lowest error delay compared to non-overlapping windows.