Makine öğrenmesi algoritmaları ile meteorolojik parametreleri kullanarak toprak radon gazının tahmini

Radon is the natural radiation source with the highest dose of exposure among all radiation sources found on earth. It consists of the degradation of natural uranium and radium elements in rocks. Factors that shape the movement of radon include meteorological factors such as the rate of decaying of radon isotopes, the fluids that fill the pores(air, water and other gases), atmospheric pressure, soil and air temperature, wind speed, and wind direction. The aim of this study is to evaluate the effects of some meteorological factors on Radon gas using Supervised Learning Algorithms and to estimate the radon gas values according to these factors. For the study, in addition to the radon levels obtained from the Seferihisar region in hourly periods between 30 October 2006 and 04 June 2007, the measurements for the parameters of Hourly Actual Pressure(hPa), Hourly 50 cm Soil Temperature(°C), Hourly Relative Humidity(%), Hourly Temperature(°C), Hourly Wind Degree(°), Hourly Wind Speed(m/sec) and Wind Direction were obtained from the Republic of Turkey Ministry of Agriculture and Forestry, General Directorate of Meteorology. To analyze the relationship between Radon and meteorological factors affecting radon with Supervised Learning Algorithms, Multiple Linear Regression, k-Nearest Neighbor, Support Vector Machines, Regression Trees, Bagging, Random Forests, XGBoost methods have been used. To test the success of applied methods K-Fold Cross-Validation(K=5) and verification tests were performed. Specification coefficient(R^2) for comparing the performance of algorithms, Mean Squared Error(MSE), Root Mean Squared Error(RMSE), Mean Absolute Error(MAE) values were used. The best result was random forests regression when performance criteria were taken into account. This method was followed by the XGBoost and k-Nearest Neighbors algorithms, which gave very close results.