CROSS-CITY CRIME RISK PREDICTION USING UNSUPERVISED DOMAIN ADAPTATION

Abstract

Predicting crime risk is essential for both city safety and the standard of living for locals. However, it is difficult to forecast the danger of crime in cities without labelled data. Collecting high-quality labelled crime data is not easy for many places because of maintenance expenses and municipal rules. Specifically, some cities may contain a large amount of labelled data, while others may have a little amount. By learning from a city with a wealth of data, a crime prediction model for a place without labelled crime data has been developed. However, this prediction job is made more challenging by the disparity in pertinent background data across cities. In order to solve the problem of context inconsistency, this study suggests an efficient unsupervised domain adaptation model (UDAC) for crime risk prediction across cities. More precisely, for every target city grid, we first find a number of comparable source city grids. We then create auxiliary contexts for the target city based on these source city grids in order to maintain consistency between the two cities. The goal of a dense convolutional network with unsupervised domain adaptation is to concurrently learn domain-invariant features for domain adaptation and high-level representations for precise crime risk prediction. Three real-world datasets are used in comprehensive tests to confirm our model's efficacy.
A key element of the expanding discipline of data science is machine learning. Various algorithms are taught using statistical techniques to create predictions or classifications and to provide important information for this project. These insights then inform business and application decision-making, hopefully influencing important growth indicators.
This project data, sometimes referred to as training data, is used by machine learning algorithms to create a model that allows them to make judgements or predictions without explicit programming. When developing traditional algorithms to accomplish the required tasks is challenging or impractical, machine learning methods are used in a broad range of datasets.