Researchers regard crime as a social phenomenon that is influenced by several physical, social, and economic factors. Different types of crimes are said to have different motivations. Theft, for instance, is a crime that is based on opportunity, whereas murder is driven by emotion. In accordance with this, we examine how well a model can perform with only spatiotemporal information at hand when it comes to predicting a single crime. More specifically, we aim at predicting theft, as this is a crime that should be predictable using spatiotemporal information. We aim to answer the question: "How well can we predict theft using spatial and temporal features?". To answer this question, we examine the effectiveness of support vector machines, linear regression, XGBoost, Random Forest, and k-nearest neighbours, using different imbalanced techniques and hyperparameters. XGBoost showed the best results with an F1-score of 0.86.

翻译：研究者认为犯罪是一种受物理、社会和经济多重因素影响的社会现象。不同犯罪类型具有不同动机，例如盗窃属于机会型犯罪，而谋杀则受情绪驱动。基于此，我们探究仅利用时空信息预测单一犯罪类型的模型效能。具体而言，我们聚焦于盗窃预测——这类犯罪理论上可通过时空信息进行预测。我们试图回答以下问题："利用时空特征能在多大程度上预测盗窃行为？"为解答该问题，我们比较了支持向量机、线性回归、XGBoost、随机森林和k近邻等模型的效果，采用不同不平衡处理技术与超参数配置。结果表明XGBoost表现最优，其F1分数达0.86。