BINNING AND IMPROVED DEEP LEARNING FOR CRIME TRENDS PREDICTION

Abstract

Future crime can be minimized as well as identified using crime prediction. Past data is used in crime prediction and future crime is predicted by analysing data with time and location. Serial criminal cases occur rapidly in present days. With better performance, accurate future crime prediction is a challenging task. For solving Crime detection problem, data mining methods are highly useful. For crime trends prediction, a new framework is proposed in recent work. In this technique, highly influenced features are selected using improved bat optimization (IBAT). At last, for getting highly accurate crime trends and for avoiding further crime, convolution neural network algorithm is used for predicting crime trends. However data set taken for this work may have noisy data values and it may affect the crime trend prediction performance and it does not focused in existing work. For rectifying those issues, for crime trends prediction, this work introduces an improved frame work. In which first input crime data will be pre-processed using missing value imputation, binning and min - max normalization. And then significant features are selected using improved cuckoo search optimization to improve the prediction. Finally, sparse regularization for convolutional neural network (SRCNN) with rectified linear units (ReLU) in hidden layers is introduced for crime trend prediction. For ReLU’s inputs, sparseness is introduced. In learning process, ReLU’s inputs are pushed into zero. This prevents unnecessary ReLU’s output increase. Experimental results demonstrate that proposed model produces better outcome with respect to accuracy, f-measure, recall and precision for Philadelphia, Chicago, and Francisco dataset.