A spatio-temporal method for crime prediction using historical crime data and transitional zones identified from nightlight imagery

ABSTRACT

Accurate crime prediction can help allocate police resources for crime reduction and prevention. There are two popular approaches to predict criminal activities: one is based on historical crime, and the other is based on environmental variables correlated with criminal patterns. Previous research on geo-statistical modeling mainly considered one type of data in space-time domain, and few sought to blend multi-source data. In this research, we proposed a spatio-temporal Cokriging algorithm to integrate historical crime data and urban transitional zones for more accurate crime prediction. Time-series historical crime data were used as the primary variable, while urban transitional zones identified from the VIIRS nightlight imagery were used as the secondary co-variable. The algorithm has been applied to predict weekly-based street crime and hotspots in Cincinnati, Ohio. Statistical tests and Predictive Accuracy Index (PAI) and Predictive Efficiency Index (PEI) tests were used to validate predictions in comparison with those of the control group without using the co-variable. The validation results demonstrate that the proposed algorithm with historical crime data and urban transitional zones increased the correlation coefficient by 5.4% for weekdays and by 12.3% for weekends in statistical tests, and gained higher hit rates measured by PAI/PEI in the hotspots test.     

Novel evaluation metrics for sparse spatio-temporal point process hotspot predictions - a crime case study

Many physical and sociological processes are represented as discrete events in time and space. These spatio-temporal point processes are often sparse, meaning that they cannot be aggregated and treated with conventional regression models. Models based on the point process framework may be employed instead for prediction purposes. Evaluating the predictive performance of these models poses a unique challenge, as the same sparseness prevents the use of popular measures such as the root mean squared error. Statistical likelihood is a valid alternative, but this does not measure absolute performance and is therefore difficult for practitioners and researchers to interpret. Motivated by this limitation, we develop a practical toolkit of evaluation metrics for spatio-temporal point process predictions. The metrics are based around the concept of hotspots, which represent areas of high point density. In addition to measuring predictive accuracy, our evaluation toolkit considers broader aspects of predictive performance, including a characterisation of the spatial and temporal distributions of predicted hotspots and a comparison of the complementarity of different prediction methods. We demonstrate the application of our evaluation metrics using a case study of crime prediction, comparing four varied prediction methods using crime data from two different locations and multiple crime types. The results highlight a previously unseen interplay between predictive accuracy and spatio-temporal dispersion of predicted hotspots. The new evaluation framework may be applied to compare multiple prediction methods in a variety of scenarios, yielding valuable new insight into the predictive performance of point process-based prediction.     

Spatial crime distribution and prediction for sporting events using social media

ABSTRACT

Sporting events attract high volumes of people, which in turn leads to increased use of social media. In addition, research shows that sporting events may trigger violent behavior that can lead to crime. This study analyses the spatial relationships between crime occurrences, demographic, socio-economic and environmental variables, together with geo-located Twitter messages and their ‘violent’ subsets. The analysis compares basketball and hockey game days and non-game days. Moreover, this research aims to analyze crime prediction models using historical crime data as a basis and then introducing tweets and additional variables in their role as covariates of crime. First, this study investigates the spatial distribution of and correlation between crime and tweets during the same temporal periods. Feature selection models are applied in order to identify the best explanatory variables. Then, we apply localized kernel density estimation model for crime prediction during basketball and hockey games, and on non-game days. Findings from this study show that Twitter data, and a subset of violent tweets, are useful in building prediction models for the seven investigated crime types for home and away sporting events, and non-game days, with different levels of improvement.     

基于随机森林和时空核密度方法的不同周期犯罪热点预测对比

犯罪预测对于制定警务策略、实施犯罪防控具有重要意义。机器学习和核密度是2类主流犯罪热点预测方法,然而目前还鲜有研究对这2类方法在不同时间周期下的犯罪预测效果进行系统比较,本文试图对此进行补充。本文以2013-2016年5月的公共盗窃犯罪历史数据作为输入,分别对比了在接下来2周、1个月、2个月、3个月4个不同时间周期随机森林方法与基于时空邻近性的核密度方法的犯罪热点预测效果,结果发现：在各时间周期上,随机森林分类热点预测方法的面积和案件量命中率均比时空核密度方法准确性高;并且2种方法均能有效地识别犯罪热点中的高发区域,其中在较小范围较短时间内随机森林识别热点中的高发区效率更高,而在较大范围较长时间周期上时空核密度方法识别高发区更优。     

Hotspots vs. harmspots: Shifting the focus from counts to harm in the criminology of place

The prevailing pattern in much of the social sciences, including geography and criminology, relies on count data. “Hotspots” — geospatial areas with disproportionally more crime than the rest of the city — are usually identified by the number of events in these areas. Yet no attention is given to their severity, or any other weighting system of harm, despite the common-sense view that not all crimes are created equal. To illustrate the value of focusing on harm in addition to count data, we turn to a spatial analysis of crime by observing crime concentrations (hotspots) against harm concentrations (harmspots), across fifteen councils in the United Kingdom. The definition of “harm” is based on the Sentencing Guidelines for England and Wales, as each crime category (n = 415) attracts a different severity weight. Both “hotspots” and “harmspots” are defined as being at least 2 standard deviations from the mean distribution within each city: This procedure creates comparable datasets. The data suggest that half of all crime events are concentrated within 3% of all street segments in the selected councils, yet harm is even more heavily concentrated, with half of all harm located in just 1% of each council [OR = 3.49; 95% CI 3.268–3.728]. The intra-unit variance was also reduced by approximately half — from 0.75% to 0.45%. We discuss the implications of using harm, in addition to counts, for research and policy by arguing that a shift in focus is required both for the development of theories and for cost-effective prevention strategies.     

The space/time behaviour of dwelling burglars: Finding near repeat patterns in serial offender data

Whilst analysis of crime for tactical and strategic reasons within the criminal justice arena has now become an established need, predictive analysis of crime remains, and probably always will be, a goal to be desired. Opening a window on this over the last 2 decades, prominent research from academia has focused on the phenomenon of repeat victimisation and more recently ‘near repeat’ victimisation, both firmly grounded in the geography of crime. Somewhat limited to the establishment of near repeat behavioural patterns in whole area data, these can be utilised for crime prevention responses on a local scale. Research reported here however, explores the phenomenon through the examination of serial offending by individual offenders to establish if such spatio-temporal patterns are apparent in the spatial behavioural patterns of the individual burglar, and if so how they may be defined and therefore utilised on a micro rather than macro scale. It is hypothesised that offenders' responsible for more than one series of offences will display consistency across their crime series within time and distance parameters for their closest offences in space. Results improve upon current knowledge concerning near repeat offending being the actions of common offenders. Testing of the extracted data indicates that offenders maintain personal boundaries of ‘closeness’ in time and space even when actions are separated by significant time spans, creating stylised behavioural signatures appertaining to their use of and movement through space when offending.     

基于时变权重组合与贝叶斯修正的中长期径流预报

高精度的中长期径流预报信息是水资源规划管理与水利工程经济运行的重要基础支撑。论文在组合预报与误差修正2类径流预报后处理方法串联应用的技术框架下,考虑径流的高度非平稳与非线性等特征,提出了基于时变权重组合和贝叶斯修正的中长期径流预报方法。应用该方法开展了云南龙江水库年、月入库径流预报的实例研究,结果表明时变权重组合平衡了已建立的随机森林与支持向量机模型在建模期与检验期预报性能的差异,经贝叶斯修正后的预报精度接近或优于两阶段各自的最优单一模型。根据年径流预报结果判断水文年型的正确率达到77.2%,月预报径流的确定性系数超过0.90。因此,该方法在提升中长期径流预报精度方面具有积极效果。     

Spatial conjunctive analysis of (crime) case configurations: Using Monte Carlo methods for significance testing

While research has repeatedly demonstrated how spatial distributions of crime can be shaped by the presence of facilities such as bars and public transport hubs, the joint influence of different facility types has rarely been explored. Spatial conjunctive analysis of case configurations (also known as qualitative comparative analysis) offers a means to identify the combinations of facility types that are most commonly found around crime events, and has been used in a small number of studies focusing on street robbery. This study extends this limited evidence base by implementing a significance test based on the Monte Carlo method using street robbery data for Austin, Texas. The results show that some of the top-ranking facility type combinations had observed frequencies that were not significantly greater than chance expectations. The accurate identification of the highest-risk environments has important implications for crime prevention.     

基于Meta-Gaussian模型的中国农业干旱预测研究

在全球气候变化背景下,干旱愈加频发,有效且可靠的农业干旱预测对于保障粮食安全和水资源安全具有重要意义。以标准化降水指数（SPI）和联合标准化土壤湿度指数（JSSI）分别表征气象干旱和农业干旱,以前期的气象干旱和农业干旱指数作为预测因子,在1~3个月预见期下基于Meta-Gaussian（MG）模型对中国1961—2015年6—8月的农业干旱进行预测,并采用Brier Skill Score（BSS）和纳什效率系数（NSE）评价MG模型的预测性能。结果表明：① 将1个月、3个月、6个月、9个月和12个月时间尺度的标准化土壤湿度指数（SSI）结合起来得到的JSSI能够对中国农业干旱的综合状况进行客观评价。② 以中国2010年和2014年遭受严重的干旱事件为例,预见期为1~3个月时,除新疆南部、青海西部以及内蒙古西部等沙漠地区外,MG模型对6—8月农业干旱预测结果的分布范围与实际干旱的分布区域较吻合,预见期越短,吻合越好。③ 预见期为1个月时,6—8月BSS ≥ 0.5的面积比例分别为0.714、0.642和0.640,NSE ≥ 0.5的面积比例分别为0.903、0.829和0.837,表明MG模型能够对中国大部分区域的农业干旱作出可靠的预测。本文结果可为中国农业干旱的监测、预警及干旱决策提供科学指导。     

陇中黄土丘陵沟壑区人工草地土壤水蚀预测模型

针对陇中黄土丘陵沟壑区土壤水蚀过程复杂且难以有效预测的问题,以定西市安家沟水土保持试验站2005—2016年1~12月人工草地径流场试验数据为主要来源,将流域月降雨量、月侵蚀性降雨量、月径流量、月降雨强度、径流场面积、径流场坡度、土壤砂粒含量、土壤粘粒含量8个因子作为输入因子,月土壤水蚀量作为输出,运用偏最小二乘法(Partial Least-Squares Regression,PLSR)和长短期记忆(Long Short-Term Memory,LSTM)循环神经网络建立人工草地土壤水蚀预测模型,并利用BP(Back Propagation)、RNN(Recurrent Neural Network)、LSTM常见神经网络模型,对模型的有效性进行评估。结果表明：PLSR将模型8个输入因子减少为4个,从而有效解决LSTM神经网络模型对样本数量要求过高的问题; PLSR和LSTM神经网络模型的结合可以有效提高模型对人工草地土壤水蚀过程的预测精度和收敛速度,预测结果的平均相对误差小于4%,相关系数高于其他3种神经网络模型,而迭代次数、均方根误差和平均绝对误差均低于其他3种模型;研究发现坡度对人工草地土壤水蚀过程影响较为明显,降雨量小于25 mm时,人工草地土壤水蚀量不会随坡度增加而明显增长,但当降雨量超过25 mm时,人工草地土壤水蚀量会随坡度明显增加。 PLSR LSTM神经网络土壤水蚀预测模型可以准确预测陇中黄土丘陵沟壑区人工草地土壤水蚀量,为该地区水土流失的准确预报提供新的思路和方法。     

Mapping crime harm and priority locations in New Zealand: A comparison of spatial analysis methods

Despite progress being advanced with spatial approaches to crime and crime control, the geography of crime harm has to date received little attention. The recent development of “Crime Harm Indices”, which weight crimes by an estimate of the relative harm they cause, offers an opportunity to improve on volume based spatial analysis approaches to identify where crime harm concentrates.This study aims to address this issue via the use of a Crime Harm Index (CHI) developed for New Zealand. By contrast to localized ‘harm-spotting’ analysis, we apply a census unit based approach to identify, at a macro level, the neighborhoods and wider communities suffering the highest crime harm in New Zealand. This approach enables harm to be viewed not only as a total Index but as a rate controlled for population and allows for the identification of census based sociodemographic factors which predict harm. Specifically, this paper compares the CHI with the New Zealand Priority Locations Index (PLI), an existing census unit based crime analysis tool which combines crime and demographic variables to identify communities vulnerable to crime and disorder issues.In this study CHI and PLI scores were calculated for Census Area Units (normally containing 3000–5000 population) across New Zealand. Bivariate correlations and a general linear model were used to determine the relationships between the CHI and PLI and additional population related variables. The CHI and PLI were weakly correlated, with population size and urban/rural categorization also accounting for CHI variance. Mapping techniques are used to illustrate outlier locations where the CHI and PLI differ widely and to identify location features which may assist in explaining CHI/PLI differences.This work exemplifies a novel geographic approach to the problem of crime harm with implications for resource allocation at national through to local levels. Wider implications for the theory and practice of crime and crime harm control are discussed, along with limitations of the study and areas for further research.     

Analysis of geomagnetic variations in south-western Nigeria

Geomagnetic variations, observed at 11 sites in south-western Nigeria, have been analysed to derive interstation transfer functions with the site at Ile-Ife as reference station. The study involves frequencies from 1 to 6 c.p.h. The reference station Ile-Ife is 160 km northward of the continental slope off the Nigerian coastline and 400 km southward of the dip equator. The analysis has been carried out separately with selected data sections of a few hours length during daytime and during the night. Thus expressed linear relations between field components are in the case of day events of dual implication: (1) for the source field structure of the equatorial electrojet, (2) for internal conductivity conditions, including the coast effect from the Bight of Benin. Conductivity anomalies are the sole cause for an observed spatial variability of night events. A 2-D thin-sheet conductivity model has been derived taking both the source and the coast effect into consideration. This model provides a reasonably good fit between observed and computed transfer functions during day and night.     

Incorporating Spatial Dependence in Predictive Vegetation Models: Residual Interpolation Methods*

Predictive vegetation modeling can be used statistically to relate the distribution of vegetation across a landscape as a function of important environmental variables. Often these models are developed without considering the spatial pattern that is inherent in biogeographical data, resulting from either biotic processes or missing or misspecified environmental variables. Including spatial dependence explicitly in a predictive model can be an efficient way to improve model accuracy with the available data. In this study, model residuals were interpolated and added to model predictions, and the resulting prediction accuracies were assessed. Adding kriged residuals improved model accuracy more often than adding simulated residuals, although some alliances showed no improvement or worse accuracy when residuals were added. In general, the prediction accuracies that were not increased by adding kriged residuals were either rare in the sample or had high nonspatial model accuracy. Regression interpolation methods can be an important addition to current tools used in predictive vegetation models as they allow observations that are predicted well by environmental variables to be left alone, while adjusting over‐ and underpredicted observations based on local factors.     

Crime concentrations and similarities in spatial crime patterns in a Brazilian context

Research within the geography of crime and spatial criminology literature most often show that crime is highly concentrated in particular places. Moreover, a subset of this literature has shown that the spatial patterns of these concentrations are different across crime types. This raises questions regarding the appropriateness of aggregating crime types (property and violent crime, for example) when the underlying spatial pattern is of interest. In this paper, using crime data from Campinas, Brazil, we investigate the crime concentrations and the similarities among different crime types across space. Similar to some recent research in another context, we find that crime is highly concentrated in Campinas but the ability to aggregate similar crime types at the street segment level is not generalizability when compared to a North American context.     

类比合成方法在干旱区内陆河径流量预报中的应用

类比合成算法是一种多维模式搜索法，它具有适用范围广、对资料要求低等优点，可用于单变量及多变量时间序列的延拓预测。通过介绍类比合成算法，并把它应用于塔里木河源流叶尔羌河、和田河年、月径流量预报，其中重点分析了模式长度和合成预报的模式个数等因素对预报结果的影响。通过实测径流资料对预报结果的检验和分析表明，类比合成算法可以较好地挖掘径流序列中隐藏的信息，在中长期水文预报中是一种行之有效的计算方法。     

Exploring spatiotemporal clusters based on extended kernel estimation methods

We examined three different ways to integrate spatial and temporal data in kernel density estimation methods (KDE) to identify space–time clusters of geographic events. Spatial data and time data are typically measured in different units along respective dimensions. Therefore, spatial KDE methods require special extensions when incorporating temporal data to detect spatiotemporal clusters of geographical event. In addition to a real-world data set, we applied the proposed methods to simulated data that were generated through random and normal processes to compare results of different kernel functions. The comparison is based on hit rates and values of a compactness index with considerations of both spatial and temporal attributes of the data. The results show that the spatiotemporal KDE (STKDE) can reach higher hit rates while keeping identified hotspots compact. The implementation of these STKDE methods is tested using the 2012 crime event data in Akron, Ohio, as an example. The results show that STKDE methods reveal new perspectives from the data that go beyond what can be extracted by using the conventional spatial KDE.     

1962-2011年长江流域极端气温事件分析

根据1962-2011 年长江流域115 个气象站点的逐日最高气温、日最低气温资料,利用线性倾向估计法、主成分分析及相关分析法,并根据选取的16 个极端气温指标,分析了该地区极端气温的时间变化趋势和空间分布规律。结果表明:(1) 冷昼日数、冷夜日数、冰冻日数、霜冻日数、冷持续日数分别以-0.84、-2.78、-0.48、-3.29、-0.67 d·(10a)^-1的趋势减小,而暖昼日数、暖夜日数、夏季日数、热夜日数、暖持续日数、生物生长季以2.24、2.86、2.93、1.80、0.83 、2.30 d·(10a)^-1的趋势增加,日最高(低) 气温的极低值、日最高(低) 气温的极高值和极端气温日较差的倾向率分别为0.33、0.47、0.16、0.19、-0.07 ℃·(10a)^-1;(2) 冷指数(冷夜日数、日最高气温的极低值、日最低气温的极低值)的变暖幅度明显大于暖指数(暖夜日数、日最高气温的极高值、日最低气温的极高值),夜指数(暖夜日数、冷夜日数) 的变暖幅度明显大于昼指数(暖昼日数、冷昼日数);(3) 空间分布上,长江上游区域冷指数的平均值大于其中下游区域,而暖指数和生物生长季则是中下游多年平均值大于上游区域(暖持续日数除外);(4) 因子分析的结果表明,除了极端气温日较差之外,各极端气温指数之间均呈现很好的相关性。     

On Blind Tests and Spatial Prediction Models

This contribution discusses the usage of blind tests, BT, to cross-validate and interpret the results of predictions by statistical models applied to spatial databases. Models such as Bayesian probability, empirical likelihood ratio, fuzzy sets, or neural networks were and are being applied to identify areas likely to contain events such as undiscovered mineral resources, zones of high natural hazard, or sites with high potential environmental impact. By processing the information in a spatial database, the models establish the relationships between the distribution of known events and their contextual settings, described by both thematic and continuous data layers. The relationships are to locate situations where similar events are likely to occur. Maps of predicted relative resource potential or of relative hazard/impact levels are generated. They consist of relative values that need careful quantitative scrutiny to be interpreted for taking decisions on further action in exploration or on hazard/impact mitigation and avoidance. The only meaning of such relative values is their rank. Obviously, to assess the reliability of the predicted ranks, tests are indispensable. This is also a consequence of the impracticality of waiting for the future to reveal the goodness of our prediction. During the past decade only a few attempts have been made by some researchers to cross-validate the results of spatial predictions. Furthermore, assumptions and applications of cross-validations differ considerably in a number of recent case studies. A perspective for all such experiments is provided using two specific examples, one in mineral exploration and the other in landslide hazard, to answer the fundamental question: how good is my prediction?     

Street profile analysis: A new method for mapping crime on major roadways

Street profile analysis is a new method for analyzing temporal and spatial crime patterns along major roadways in metropolitan areas. This crime mapping technique allows for the identification of crime patterns along these street segments. These are linear spaces where aggregate crime patterns merge with crime attractors/generators and human movement to demonstrate how directionality is embedded in city infrastructures. Visually presenting the interplay between these criminological concepts and land use can improve police crime management strategies. This research presents how this crime mapping technique can be applied to a major roadway in Burnaby, Canada. This technique is contrasted with other crime mapping methods to demonstrate the utility of this approach when analyzing the rate and velocity of crime patterns overtime and in space.     

The concentration of urban crime in space by race: evidence from South Africa

Crime inequality in neighborhoods by race is blamed on social inequalities borne out of segregation and economic discrimination. South Africa is a country synonymous with racial-spatial segregation and discrimination as a result of legislatively enforced policies of the former apartheid government. This study examines whether urban crime inequalities by race exist in the city of Tshwane, South Africa and identifies the empirical causes of these crime inequalities. Violent and sexual crime was found to concentrate in Black African neighborhoods, while property crime was concentrated in neighborhoods classified as “Mixed”. The causes of crime in neighborhoods were found to vary across racial groups with results suggesting non-uniformity in the extent to which the various constructs impact crime based on race. The results challenge the notion that segregation and economic discrimination uniformly impacts affected communities. Explanations for the findings are provided in the context of an increasingly eclectic post-apartheid South African city.