Development of a Predictive Model of Burgling to Homes based on Machine Learning

The general objective of the paper is to determine the probability of burglary in the city of Chennai, and determine the factors that influence this type of crime, thus identifying the houses most prone to it. For this, the aforementioned public database of police cases is used to design and execute a python program based on neural networks that manages to obtain predictions of house robberies in different areas of the city. This paper investigates the relevant features in the obtained data sources to determine which are most important for theft prediction. Once data is selected, Data preprocessing is performed to prepare the extracted data is prepared and cleaned so that it can be correctly included in subsequent models. In the case of contextual information from images, convolutional neural networks are used for feature extraction. Then, neural network architecture is built with one input for each information source and a concatenation layer to merge the inputs later. Finally, a classifier is trained using fully connected layers and an output layer with a sigmoid activation function. With these probabilities, the classification of each record is finally obtained and, therefore, the prediction of house burglaries. The result of the neural network built with a classic machine learning model such as SVM is compared. Each model's hyper-parameters are tuned to get improvements in each model's performance. The satisfaction of this objective will be determined according to the feasibility of building this model, that is, if probabilities can actually be obtained using the aforementioned data sources. The model will be evaluated through different metrics such as accuracy, precision, recall and F1 score.