Imputation by feature importance (IBFI): A methodology to envelop machine learning method for imputing missing patterns in time series data.

A new methodology, imputation by feature importance (IBFI), is studied that can be applied to any machine learning method to efficiently fill in any missing or irregularly sampled data. It applies to data missing completely at random (MCAR), missing not at random (MNAR), and missing at random (MAR). IBFI utilizes the feature importance and iteratively imputes missing values using any base learning algorithm. For this work, IBFI is tested on soil radon gas concentration (SRGC) data. XGBoost is used as the learning algorithm and missing data are simulated using R for different missingness scenarios. IBFI is based on the physically meaningful assumption that SRGC depends upon environmental parameters such as temperature and relative humidity. This assumption leads to a model obtained from the complete multivariate series where the controls are available by taking the attribute of interest as a response variable. IBFI is tested against other frequently used imputation methods, namely mean, median, mode, predictive mean matching (PMM), and hot-deck procedures. The performance of the different imputation methods was assessed using root mean squared error (RMSE), mean squared log error (MSLE), mean absolute percentage error (MAPE), percent bias (PB), and mean squared error (MSE) statistics. The imputation process requires more attention when multiple variables are missing in different samples, resulting in challenges to machine learning methods because some controls are missing. IBFI appears to have an advantage in such circumstances. For testing IBFI, Radon Time Series Data (RTS) has been used and data was collected from 1st March 2017 to the 11th of May 2018, including 4 seismic activities that have taken place during the data collection time.

Delegated Regressor, A Robust Approach for Automated Anomaly Detection in the Soil Radon Time Series Data.

We propose a new method based on the idea of delegating regressors for predicting the soil radon gas concentration (SRGC) and anomalies in radon or any other time series data. The proposed method is compared to different traditional boosting e.g., Extreme Gradient Boosting (EGB) and simple regression methods e.g., support vector regressors with linear kernel and radial kernel in terms of accurate predictions. R language has been used for the statistical analysis of radon time series (RTS) data. The results obtained show that the proposed methodology predicts SRGC more accurately when compared to different traditional boosting and regression methods. The best correlation is found between the actual and predicted radon concentration for window size of 2 i.e., two days before and after the start of seismic activities. RTS data was collected from 05 February 2017 to 16 February 2018, including 7 seismic events recorded during the study period. Findings of study show that the proposed methodology predicts the SRGC with more precision, for all the window sizes, by overlapping predicted with the actual radon time series concentrations.

Descriptive analysis and earthquake prediction using boxplot interpretation of soil radon time series data.

Correlation of radon anomalies with meteorological parameters and earthquake occurrence has been reported in many studies. This paper reports descriptive statistical analysis and boxplot contingent earthquake prediction based upon soil radon time series data. Data has been collected over a fault line, passing beneath the Muzaffarabad, for the period of one year. Soil radon gas (SRG) was measured using RTM 1688-2 radiometric instrument (made by SARAD GmbH). The range of radon in soil was found to be 14349 Bqm-3, whereas the ranges of temperature, pressure and relative humidity were found as 38.50 C0, 29 mbar and 67% respectively. SRG data shows that time series follows normal distribution. Values of coefficient of variation (CV) indicate the consistency of the recorded values of radon in soil and metrological parameters. Variance inflation factor (VIF) and Durbin Watson test (d) indicate a moderate multicollinearity and autocorrelation between variables. The analysis of radon time series using boxplots and meteorological parameters show specific patterns in radon concentrations (outliers, variant IQRs, first quartile values, and median values) due to pre-earthquake underground seismic activities. On the basis of these patterns earthquake may be more early predicted without using complicated predictive systems. Boxplots also predicted that there is no significant pattern found in dispersion of meteorological factors measured in this study. To the best of our knowledge this is first ever attempt to predict earthquake using boxplot explanation.

Experimental study of the effect of water level and wind speed on radon exhalation of uranium tailings from heap leaching uranium mines.

Water level and wind speed have important influences on radon release in particle-packing emanation media. Based on radon migration theory in porous media under three water level conditions, an experimental setup was designed to measure the surface radon exhalation rate of uranium tailings from heap leaching uranium mine at different water levels and wind speeds. When the water level was at 0.3 m (overlying depth 0.05 m), radon transfer velocities at the gas-liquid interface were also measured at different wind speeds. Results show that when the water level was equal to or lower than the surface of the sample, the radon exhalation rate increased with increasing wind speed and decreased with increasing water level. When the water level was higher than the surface of the sample, radon exhalation rate of the water surface increased with increasing surface wind speed. The wind speed, however, was less influential on the radon exhalation rate as the depth of the overlying water increased, with a dramatic decrease in radon release. That said, at different wind speeds, radon transfer velocities at the gas-liquid interface were consistent with literature. On the other hand, changes in wind speed had significant influences on the radon transfer velocity at the gas-liquid interface, with the effect less pronounced at higher wind speeds.

Automated anomalous behaviour detection in soil radon gas prior to earthquakes using computational intelligence techniques.

In this article, three computational intelligence (CI) models were developed to automatically detect anomalous behaviour in soil radon gas (222Rn) time series data. Data were obtained at a fault line and analysed using three machine learning techniques with the aim at identifying anomalies in temporal radon data prompted by seismic events. Radon concentrations were modelled with corresponding meteorological and statistical parameters. This leads to the estimation of soil radon gas without and with meteorological parameters. The comparison between computed radon concentration and actual radon concentrations was used in finding radon anomaly based upon automated system. The anomaly in radon time series data could be considered due to noise or seismic activity. Findings of study show that under meticulously characterized environments, on exclusion of noise contribution, seismic activity is responsible for anomalous behaviour seen in radon time series data.

A method for determining the analytical form of a radionuclide depth distribution using multiple gamma spectrometry measurements.

When characterizing environmental radioactivity, whether in the soil or within concrete building structures undergoing remediation or decommissioning, it is highly desirable to know the radionuclide depth distribution. This is typically modeled using continuous analytical expressions, whose forms are believed to best represent the true source distributions. In situ gamma ray spectroscopic measurements are combined with these models to fully describe the source. Currently, the choice of analytical expressions is based upon prior experimental core sampling results at similar locations, any known site history, or radionuclide transport models. This paper presents a method, employing multiple in situ measurements at a single site, for determining the analytical form that best represents the true depth distribution present. The measurements can be made using a variety of geometries, each of which has a different sensitivity variation with source spatial distribution. Using non-linear least squares numerical optimization methods, the results can be fit to a collection of analytical models and the parameters of each model determined. The analytical expression that results in the fit with the lowest residual is selected as the most accurate representation. A cursory examination is made of the effects of measurement errors on the method.