No abstract

Radon is a chemically inert, tasteless, and odorless gas, which causes lung cancer in people who are exposed to higher concentrations for extended periods of time. It is a byproduct of the decay of uranium in the soil. High concentrations are present in closed units, like houses, schools, etc. To identify houses with unacceptably high radon levels, the radon concentration for each zip code in Ohio needs to be measured. However, not all of the zip codes are surveyed, owing to reasons such as inapproachability. In places where data is unavailable, the concentrations need to be estimated using interpolation techniques. This article proposes the application of knowledge‐based neural network approaches, namely prior knowledge input method, source difference method, and space‐mapped neural network method for modeling and predicting radon concentrations. To this end, knowledge available in the form of uranium concentration data is exploited. Modeling accuracies of the proposed techniques are compared with those of the commonly used multilayer perceptron networks (employed in one of our recent work) and conventional interpolation techniques. © 2012 American Institute of Chemical Engineers Environ Prog, 32: 355‐364, 2013

Radon (Rn) is a chemically inert, naturally occurring radioactive gas. It is one of the main causes of lung cancer second to smoking, and accounts for about 25,000 deaths every year in the US alone according to the National Cancer Institute. In order to initiate preventive measures to reduce the deaths caused by radon inhalation, it is helpful to have radon concentration data for each locality, e.g. zip code. However, such data are not available for each and every zip code in Ohio, owing to several reasons including inapproachability. In places where data is unavailable, radon concentrations must be estimated using interpolation techniques. This paper presents a new interpolation technique based on Artificial Neural Networks (ANNs) for modeling and predicting radon concentrations in Ohio, US. Several ANNs were first trained and then validated using available data. From the resulting models, the model with lowest validation error was identified. Model accuracies using the proposed approach was found to be significantly better compared to conventional interpolation techniques such as Kriging and Radial Basis Functions.