The Open Environmental & Biological Monitoring Journal, 2011, 4, 21-31 21 1875-0400/11 2011 Bentham Open Open Access Development of an ANN Interpolation Scheme for Estimating Missing Radon Concentrations in Ohio Arjun Akkala 1 , Vijay Devabhaktuni 1,* , Ashok Kumar 2 and Deepak Bhatt 1 1 EECS Department, The University of Toledo, MS 308, 2801 W. Bancroft St., Toledo, OH 43606, USA 2 Department of Civil Engineering, The University of Toledo, MS 307, 2801 W. Bancroft St., Toledo, OH 43606, USA Abstract: Radon (Rn) is a chemically inert, naturally occurring radioactive gas. It is one of the main causes of lung can- cer 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 ra- don 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 predict- ing 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. Keywords: Artificial neural networks, Interpolation, Modeling, Ohio, Radon, Zip code. 1. INTRODUCTION Radon, which is an invisible, colorless, odorless gas, is a daughter element in the radioactive decay series of uranium. Uranium is widespread in small quantities in rocks and sediments. Both radon and its decay products are radioactive. Radon can cause lung cancer in people exposed to high lev- els over a long period of time [1], a health issue that many homeowners unknowingly face. Radon is responsible for about 25,000 lung cancer deaths every year in the US. About 2,900 of these deaths occur among people who never smoked [2]. Radon is classified as a Class A carcinogen by the U.S. Environmental Protection Agency [3]. The USEPA and other organizations have launched research efforts to help assess risks and remedial options. In this context, some of the questions worth investigation include: (i) What is the statistical and spatial distribution of indoor radon; (ii) What methods can be used to reduce radon concentrations in homes; (iii) What is the risk as a function of exposure; etc. There have been ongoing efforts, including those at the Uni- versity of Toledo, in terms of maintaining radon concentra- tion databases for states with high radon levels, e.g. Ohio. Although Ohio’s radon concentrations are not as high as those in some other states, they are well above the U.S. na- tional average. With an objective of providing a healthy living environ- ment, the USEPA continues to support preventive actions for *Address correspondence to this author at the EECS Department, The Uni- versity of Toledo, MS 308, 2801 W. Bancroft St., Toledo, OH 43606, USA; Tel: (419) 530-8172; Fax: (419) 530-8146; E-mail: Vijay.Devabhaktuni@utoledo.edu all homes with higher radon activity. For instance, Ohio De- partment of Health (ODH) runs a campaign aimed at measur- ing radon concentrations across Ohio. Health authorities, in conjunction with county health departments, commercial testing services, and university researchers have so far gath- ered information for more than 130,000 homes/schools across Ohio [4-8]. Data management has been carried out using different database management systems [9-13]. Radon data is unavailable for some locations or zip codes, owing to several reasons including inapproachability. Fig. (1) shows the geometric mean of concentrations across Ohio (based on the Ohio Radon Information System at the University of Toledo). The regions marked in white color correspond to the regions of Ohio, for which, no data are available. For zip codes where multiple data are available, a general practice has been to compute the Geometric Mean (GM) of all available data to account for random data collec- tion by homeowners. The current database has radon concen- trations available for 1261 zip codes out of 1492 zip codes in Ohio. For regions with no data availability, radon concentra- tions need to be estimated using interpolation techniques. In this work, we propose a new ANN based scheme for model- ing and predicting radon concentrations. Neural networks employed in this work are 3-layer multi-layer perceptrons often referred to as 3-layer MLP or simply MLP3. The distribution of radon concentrations depicted in Fig. (1) compares well with the general distribution of uranium across Ohio. As well, the results are in line with the general geology observed by Harrell et al. [4, 5].