Optimized Features and Deep Learning Based Crime Trends Prediction

Crime Prediction is an effort of determining future crime with the intention of diminishing them. Post-analysis of the data from past events, Crime prediction forecasts the future crime on the basis of time and location. In recent decades, the rapidly increasing series of criminal cases make accurate future crime prediction a difficult task. For crime trend prediction, a Prophet Model and Keras stateful LSTM has utilized in a recent study, in which missing value imputation and feature selection has carried out by simple computation as regards the enhancement of dataset quality and crime prediction. Though this method is simple to be applied, yet it has the inadequacy to prove its prediction accuracy. Because the sequence form of the input existed in LSTM maximizes the complexity while estimating the impact of each variable. For confronting these challenges, this study proposes a novel approach explicitly for crime trends prediction. In this model, the missing values have replaced by first Predictive Mean Matching method in the dataset. In addition, an improved bat optimization (IBAT) takes place to extract highly subjective features from the dataset, concerning the performance enhancement of trends forecasting and reduction of time computation. Ultimately, the convolution neural network algorithm has further involved predicting the crimes trends in order to reduce future crime.