Anomaly detection in atomic clocks using machine learning and time series approach

Atomic clocks (ACs) are the most vital devices used by national standard agencies of a nation in order to maintain a continuous and stable time scale.Errors in the prediction of time by the ACs are caused by external factors called anomalies.Monitoring atomic clocks and detecting anomalies is essential for them to serve their purpose.The present study proposes a machine learning-based time-series approach for detecting anomalies in Indian AC data.A point anomaly has been simulated in the AC data, and a prediction has been made.The anomalies have been inserted into the 52-day phase data from August-September 2019 at the National Physical Laboratory, New Delhi.Development has been done in Jupyter Notebook using Python.Based on the analysis, the outcomes showed that precision was achieved at 0.9643, accuracy at 0.9310, recall score was 0.9643, and F1-score was 0.9643, denoting that the obtained results are higher in the 7th and 8th clock, which proves that the developed model is a good fit.