COMPARISON OF THREE DISTRIBUTION FREE CLASSIFICATION TECHNIQUES APPLIED TO CRIME DATA OF NIGERIA PRIOR AND POST COVID-19 PANDEMIC

Abstract

This paper considers a comparison and evaluation of the performance of the three distribution-free classification methods in classifying states in Nigeria. The methods were CT, KNN, and ANN. The methods classified the state into high and low crime rates using selected variables impacting crime rates. The classification results showed that the CT performed best, correctly classifying 8 as high and misclassifying 3, which yields an apparent error of 27.27%, and also correctly classifying 12 as low and misclassifying 2, which gives an apparent error of 14.29%, 80% accuracy, 80% sensitivity, and 80% specifity for training sample. While for the testing sample, the CT correctly classified 3 as high and misclassified 1, which yields an apparent error of 25%; it correctly classified 6 as low and misclassified 2, which gives an apparent error of 25%, 75% accuracy, 60% specifity, and 75.7% sensitivity, as shown in Table 7, respectively. The KNN method resulted in an apparent error of 66.67%, an accuracy of 41.67%, 42.86% sensitivity, and 40% specifity for testing data. While for training, in Table 3 below, KNN has an apparent error of 66.67%, an accuracy of 88%, 90% specifity, and 86.67% sensitivity, respectively. Lastly, the ANN did not perform well; correctly classified gives an apparent error of 100%, an accuracy of 0%, 0% sensitivity, and 0% specificity in the training sample, while for the testing sample, the method has an accuracy of 50%, 28.57% sensitivity, and 100% specificity. However, it offers many advantages that make it a useful method.