Abstract:A crime is a punishable offence that is harmful for an individual and his society. It is obvious to comprehend the patterns of criminal activity to prevent them. Research can help society to prevent and solve crime activates. Study shows that only 10 percent offenders commits 50 percent of the total offences. The enforcement team can respond faster if they have early information and pre-knowledge about crime activities of the different points of a city. In this paper, supervised learning technique is used to predict crimes with better accuracy. The proposed system predicts crimes by analyzing data-set that contains records of previously committed crimes and their patterns. The system stands on two main algorithms - i) decision tree, and ii) k-nearest neighbor. Random Forest algorithm and Adaboost are used to increase the accuracy of the prediction. Finally, oversampling is used for better accuracy. The proposed system is feed with a criminal-activity data set of twelve years of San Francisco city.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to use spatio - temporal data to predict criminal activities, in order to help law enforcement departments prevent and respond to crimes more effectively. Specifically, the authors used supervised learning techniques to improve the accuracy of crime prediction. The following are the main problems and goals of this research: 1. **Understanding crime patterns**: - Researchers hope to find patterns among the time, location and type of criminal activities by analyzing the crime data set of San Francisco in the past 12 years. - These patterns can help identify crime hot - spot areas (i.e., places where crimes occur frequently), thus providing valuable information for law enforcement departments. 2. **Improving prediction accuracy**: - In order to improve the prediction accuracy, researchers adopted a variety of machine - learning algorithms, including Decision Tree, K - Nearest Neighbor (KNN), Random Forest and AdaBoost, etc. - These algorithms process data in different ways to find the most effective prediction model. 3. **Handling the class imbalance problem**: - The frequencies of different types of crimes in the data set vary greatly, resulting in the class imbalance problem. For example, some crime types are very common, while others are very rare. - To solve this problem, researchers used oversampling and undersampling methods, especially SMOTE (Synthetic Minority Over - sampling Technique) and random undersampling. 4. **Providing practical prediction tools**: - The ultimate goal is to develop a crime prediction system that can be actually applied to law enforcement departments, enabling them to know potential criminal activities in advance and take corresponding preventive measures. In summary, this paper aims to improve the accuracy of crime prediction by analyzing spatio - temporal data and using machine - learning algorithms, and solve the class imbalance problem in the data, so as to help law enforcement departments prevent crimes more effectively.

Crime Prediction Using Spatio-Temporal Data

DeepOffense: a Recurrent Network Based Approach for Crime Prediction

Spatio-temporal prediction of crimes using network analytic approach

Crime Prediction Based On Crime Types And Using Spatial And Temporal Criminal Hotspots

Spatio-Temporal Analysis of Urban Crime Leveraging Multisource Crowdsensed Data

Spatiotemporal Crime Hotspots Analysis And Crime Occurrence Prediction

Predicting Crime Using Spatial Features

Modeling Temporal-Spatial Correlations for Crime Prediction.

Crime Prediction Using Multiple-ANFIS Architecture and Spatiotemporal Data

Multi-Type Urban Crime Prediction.

Crime Prediction using Machine Learning with a Novel Crime Dataset

Exploring Spatio-Temporal and Cross-Type Correlations for Crime Prediction

A conditional machine learning classification approach for spatio-temporal risk assessment of crime data

ESCORT: Fine-Grained Urban Crime Risk Inference Leveraging Heterogeneous Open Data

Big Data Analytics and Mining for Crime Data Analysis, Visualization and Prediction.

Crime Prediction with Historical Crime and Movement Data of Potential Offenders Using a Spatio-Temporal Cokriging Method

A Comparative Study on Crime in Denver City Based on Machine Learning and Data Mining

A Comparative Analysis of Multiple Methods for Predicting a Specific Type of Crime in the City of Chicago

To be or not to be? A spatial predictive crime model for Rochester

CrimeTensor : Fine-Scale Crime Prediction via Tensor Learning with Spatiotemporal Consistency

Time-series Crime Prediction Across the United States Based on Socioeconomic and Political Factors