Smart Policing is the need of the hour as exponentially increase in crime rates is observed across the country, to intelligently deploying forces / rescue resources in preventing comes from happening and to minimize the effect of crimes. A crime is defined as a deliberate act that results in bodily or psychological harm as well as property damage or loss. Law enforcing agencies strive to come up with efficient prevention measures since the amount and variety of criminal acts are growing at an alarming rate. As technology scales up, the criminal activities are absolutely alarming, leaving the conventional criminal identification techniques incompetent. Today, the Police personnel across each country immensely reliant on digital information such as images, emails, video and etc. of crime scenes to accelerate investigation. It is urgent to devise technologies to achieve better system modelling that use machine learning algorithms in order to assist for easy and quick prediction of crimes. This article proposes a model that has convolutional neural networks, Long Short Term Memory (LSTM) and Multiplicative Attention. Simulation results of the proposed idea exhibit impressive prediction when conventional models such as Deep Learning for Real-Time Crime Forecasting (DLRTCF), Linear Regression (LR), Additive Regression (AR) and Decision Stump (DS) respectively. The prediction occurs at an average processing time between 1150ms and 2118ms which is highly significant.