
Machine learning methods as a tool to analyse incomplete or irregularly sampled radon time series dataMachine learning methods as a tool to analyse incomplete or irregularly sampled radon time series data 
"/ミロソラフ, ヤニック/"ミロソラフ, ヤニック ,
"/Bossew, Peter/"Bossew, Peter ,
"/Kurihara, Osamu/"Kurihara, Osamu
内容記述
Identifying factors which control the temporal dynamics of physical quantities is a recurrent problem in physics. One example is time series of radon concentration in the air, which is influenced by environmental variable such as temperature, humidity, atmospheric stability, and so on. Understanding the behavior of radon variability in the environment over longperiod is necessary in terms of the indoor radon concentration, which is regulated in e.g. the European Union and the US. This presentation provides an example of the application of five machine learning methods to the indoor radon concentration data obtained at authors’ institute for the period between 2011 and 2016. These methods were employed to reveal the factors influencing the radon dynamics among various meteorological quantities. The performance of each method was evaluated using six statistical metrics, namely the root mean square error (RMSE), the mean absolute error (MAE), the index of agreement (IA), the fractional bias (FB), ratio (RI) and adjusted coefficient of determination (R2adj).As a result, it turned out that the Random Forest method was superior to the other methods. More than 80% of the indoor radon concentration values could be explained by this method as a function of temperature, relatively humidity and day of the year. In comparison, only 35% of the values were explained by a conventional multiple regression analysis using eight predictor quantities.
The Third EastEuropean Radon Symposium (TEERAS 2017) における発表