@misc{oai:repo.qst.go.jp:00066931,
 author = {ミロソラフ, ヤニック and Bossew, Peter and Kurihara, Osamu and ミロソラフ ヤニック and 栗原 治},
 month = {Sep},
 note = {Anomaly detection, in data mining, is a process to identification of events that do not match the overall or “background” pattern of items presents in a dataset. Anomalous observations can principally have two origins: 1) A true anomalous process of the observed quantity, distinct from the “background process”, caused by anomalous changes of controlling environmental quantities; 2) A perturbation of the observation process (sampling, measurement), due to device malfunction, statistical outliers, or unplanned response of the device to environmental conditions. 
Anomaly detection is a hot topic in radon measurement, especially in relation to geospatial analysis of Rn response to seismic activity. In this work, we first discuss shortly the concept of anomaly vs. background. Second, machine learning algorithms and statistical methods were implemented to detecting and classifying anomalies in long-term data series. The study was performed using over 2 years’ data of continuous monitoring of radon, thoron and CO2 concentration in soil gas collected at the QST/NIRS site, Chiba, Japan. The metrological and seismic data were obtained from the Japan Metrological Agency., 14th International Workshop ? GARRM (GEOLOGICAL ASPECTS OF RADON RISK MAPPING)における発表},
 title = {Anomaly detection in the long-term data of radon and thoron measurement in soil using machine learning methods},
 year = {2018}
}