Crime Feature Selection Constructing Weighted Spanning Tree

Abstract

The proposed work demonstrates a rough set based feature selection scheme for selecting crime features from online newspaper reports of crime performed against women in India. Only the verbs present in the crime reports are considered as the extracted features for crime analysis task. To select only the distinct verbs, all the words with common synonyms are identified and replaced by a single word. Most often the set of features contains the relevant as well as many irrelevant features. Hence, for any classification task, it is highly essential to select only the relevant features for accurate classification. In the proposed work, the rough set theory based relative indiscernibility relation is used to measure the similarity score between two features relative to the crime type. Then a weighted undirected graph has been generated that comprises the features as nodes and the inverse similarity score between two features as the weight of the corresponding edge. Prim’s algorithm is applied to obtain a minimal spanning tree. Finally, a feature selection algorithm has been developed that selects the highest degree node and removes it from the tree iteratively until the modified graph becomes a null graph. The selected nodes are considered as the important features sufficient for crime reports categorization.