This is simple implementation of anomaly detection based on k-means clustering. URL : http://amid.fish/anomaly-detection-with-k-means-clustering

Algorithm

1. Sampling normal data (input: time series, output: segmented time series)

2. Make starting and end points zero, multiplying window functions. (input: output of 1 , output windowed segments)

3. Learn K-means clustering model.

4. Construct reconstruction model. 4-1) sampling original data. 4-2) Find closest cluster centroid of each sampled segment. 4-3) Reconstuct data with 4-2 result.

5. Decide anomaly based on reconstruction error.

Result Example 1

Result Example 2