Introduction:

Spatiotemporal data related to the public security have been growing at an exponential rate during the recent years. In order to facilitate crime prevention, several scholars have developed models to predict crime. Most used historical crime data alone to calibrate the predictive models. The crime risk area prediction, based on the relevant influencing factors of criminal activities, refers to the correlation between criminal activities and physical environment, which both derived from the ‘‘routine activity theory’’. Traditional crime risk estimation methods usually detect crime hotspots from the historical

Results:
By comparing the prediction results of different machine learning models before and after adding covariates, the following indicators are used for evaluation. Grid Hit Rate HitRa refers to the ratio between the number of predicted correct hotspot grids and the total number of actual hotspot grids. Case Hit Rate HitRn refers to the ratio between the actual number of cases in the forecast correct hot grids and the total number of cases in the study area in this period. The larger the value of HitRn is, the more cases are included in the hot grids, and the higher the accuracy of prediction is

Conclusion:

Six machine learning algorithms are applied to predict the occurrence of crime hotspots in a town in the southeast coastal city of China. The following conclusions are drawn:

The prediction accuracies of LSTM model are better than those of the other models. It can better extract the pattern and regularity from historical crime data. The case hit rate of the LSTM model used in this paper was 59.9%, and the average grid hit rate was 57.6%, which was improved compared with the previous research results, For the future research, there are still some aspects to be improved.

Important Sentences

• Spatiotemporal data related to the public security have been growing at an exponential rate during the recent years
• According to the experimental results, we found that the prediction accuracy of the prediction accuracy of the Long Short-Term Memory (LSTM) model was improved after adding built environment covariates, and the average prediction index-HitRa of 13 experimental periods increased by percentage points increased by 12.8 percentage points, the average prediction index-HitRn of experimental periods increased by percentage points, and the average prediction index-HitEn of 13 experimental periods increased by 10.4 percentage points
• In this paper, six machine learning algorithms are applied to predict the occurrence of crime hotspots in a town in the southeast coastal city of China
• In the biweekly forecast, the highest case hit rate for the two-robbery type is 31.97%, and the highest grid hit rate is 32.95%; Liu et al Used the random forest model to predict the hot spots in multiple experiments in two weeks under the research scale of 150 m × 150 m
• The case hit rate of the LSTM model used in this paper was 59.9%, and the average grid hit rate was 57.6%, which was improved compared with the previous research results, For the future research, there are still some aspects to be improved
• Future research will assess the impact of changing grid sizes on prediction accuracy