A knowledge centric hybridized approach for crime classification incorporating deep bi-LSTM neural network

In recent years, the crime rate has increased considerably and there is a need to properly identify the different types of crimes so that it can be tackled. In this paper, a Bi-LSTM neural network for classification is proposed that classifies the different types of crime on data collected from Google News and Twitter. The data is pre-processed and an initial step of labeling is performed with the help of Fuzzy c-means algorithm and Term Frequency – Inverse Document Frequency vectors. GloVe word embeddings were performed for feature extraction. Dynamically generated ontologies with minimal human supervision using a weighted graph modeled from Google News and Social Web like Twitter has been encompassed in order to enhance the quality of crime classification. The proposed method has proven, after experiments, to achieve evaluation metrics better than the existing methods; evaluated on four different datasets and compared with four different methods with an increase in Accuracy and decrease in FNR for four distinguished datasets.