A Novel Hybrid RNN-ELM Architecture for Crime Classification

Abstract

Extreme Learning Machine (ELM) is a single layer feedforward neural network (SLFN) that has shown remarkable results in regression and classification (multi-class) problems. The theories on ELM indicates that the hidden neurons can be randomly generated. In this paper, we introduce a hybrid Recurrent Neural Network(RNN) – ELM hybrid structure for crime hotspot classification. The RNN extracts the features from the data and learns using Long Short-Term Memory (LSTM) and finally ELM is applied at the end of the layers for our classification problem. The dataset used for this study is Philadelphia’s crime data. The dataset is also tested with RNN using backpropagation without ELM. With ease of implementation, fast learning speed and better accuracy, RNN-ELM clearly outperformed RNN with backpropagation.