Sufficient Dimension Reduction Based Classification of Nigerian Cities by Crimes against Properties Safety

Abstract

This study adopts the method of Sufficient Dimension Reduction (SDR) to estimate sufficient predictors for visualizing the data of crimes against pproperties in Nigerian cities and training statical classification models that are capable of efficiently detecting true safety status of such cities without losing information. Modified version of the sliced inverse regression (SIR) methods was adopted by replacing the usual maximum likelihood covariance estimator by the Hybridized Smoothed Maximum Entropy Covariance Estimator (HSMEC) proposed by Olorede and Yahya in the dimension reduction step. All the seven statistical classifiers achieved excellent results based the first sufficient predictor estimated by the modified (SIR HSMEC) with k-Nearest Neighbour model with one optimal neighbour achiving false positive rate of 0% and 100% classification accuracy, sensitivity, specificity, and area under the curve, respectively.