Spatial Eigenvector Predictive Mapping Using a Second Order Eigenfunction Eigen-decomposition Autocorrelation Specification in a Regional Convolution Neural Network in a Geo-Intelligent Smartphone App to Identify Lead Contamination in Primary and Secondar

Abstract

Currently, there are no lead-contamination models to describe the locations of aggregation/non-aggregation sites (i.e., hot-and-cold spots) in Hillsborough County, Florida, USA. We generated a Global Moran's index (I) of spatial autocorrelation to identify hot-and-cold spots of lead contamination in Hillsborough County. The data used were based on lead sampling of water sources within the internal environments of elementary, middle, and high schools. A second-order eigenfunction eigen-decomposition algorithm embedded in a regional convolutional neural network (R-CNN) machine learning [ML] in a geo-spatial artificial intelligence [geo-AI] smartphone application yielded a model that identified Kingswood Elementary School as the most clustered geographic location with the highest lead contamination concentration levels. The spatial autocorrelation model also identified Essrig Elementary School as the least clustered geographic capture point. The Moran's I diagnostic summary plot revealed a final Z-score of 8.347 and a p-value of 0.000. Mapping hotspots of lead concentration using a second-order eigenfunction eigen-decomposition algorithm within an instantaneous R-CNN, ML, geo-AI, iOS pipeline can allow school administrative boards and policymakers to allocate resources to at-risk areas with lead contamination.