Index Mapped Ordinal Encoding Method for Federated Machine Learning in Crime Detection

Abstract

Abstract

Feature Engineering is extracting features from a raw dataset using the understanding of the domain's problem.
Thus, an improved feature extraction process can enhance the performance of machine learning algorithms. The resultant effect is the increase in the accuracy of a model in detecting new knowledge. However, the pyramid of unstructured data generated by surveillance devices and business transactions into databases is alarming and calls for serious attention. Transforming this messy data to a machine-useable format at the edge for prediction and classification exercises is therefore challenging. Although there have been some techniques for this data preprocessing phase in the model building but they are not totally spared from loss of data, duplication and difficult implementation. This is an examination of the efficacy of a novel Index Mapped Ordinal Encoding

Method

(IMOEM) for machine learning algorithm in terms of precision, recall and accuracy. The performance of IMOEMbuilt on this dataset with respect to precision, recall, f-score and accuracy results were significantly effective. The model performed exceptionally well with no loss of accuracy either in precision or recall values, especially when applied to the decision tree based Models. Data scientists are therefore encouraged to embrace the use of IMOEM.