ENSEMBLE FEATURE SELECTION AND DEEP LEARNING BASED CRIME TRENDS PREDICTION MODEL

  • J. Jeyaboopathiraja, Dr. G. Maria Priscilla
Keywords: Crime prevention, missing value imputation, pattern significant features, min - max normalization and Support vector machine.

Abstract

Very recently, crime examinationis essential to disclose the complexity in crime database and this may facilitate the person who are engage in imposition of law for catching criminals and leading the crime avoiding policies. Capacity of forecasting the upcoming crime depending on place, motive and occasion acts as a important information for the crime analyst from planned or deliberate viewpoint.however, to forecast upcoming crime exactly with enhancedaccomplishment is a tough work as the number of crime is increased now a days. Hence, crime forecasting technique is vital to discover upcoming crimes and decrease them. In recent work introduces an improved frame work for crime trends prediction. 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 establishes a sparse regularization for the convolutional neural network (SRCNN) through rectified linear units (ReLU) in CNN hidden layers. Through the introduction of deficiency in ReLU input, they are moved to null in training procedure. However, single optimization method will provides lesser optimal features and it leads to poor crime trend prediction results. And fully connected layer in the convolutional neural network provides inaccurate classification results. To overcome these problemsthis work introduces a model for crime trendforecast. Here,omitteddata in the dataset will be imputed based on fuzzy neural network and then binning and min - max normalization will be done. And then significant features are selected from the dataset based on Ensemble mutation based chicken swarm optimization algorithm (EMCSO). In which ensembling will be done by using averaging method. Finally crime trends prediction will be performed using hybrid improved convolutional neural network and support vector machine (ICNNSVM). SVM will be integrated in the fully connected layer of CNN for classification. Outcome of assessment demonstrates the success of introduced model with regard to accuracy, precision and recall.

Published
2021-08-04
How to Cite
Dr. G. Maria Priscilla, J. J. (2021). ENSEMBLE FEATURE SELECTION AND DEEP LEARNING BASED CRIME TRENDS PREDICTION MODEL. Design Engineering, 6454- 6472. Retrieved from http://thedesignengineering.com/index.php/DE/article/view/3144
Section
Articles