Crime Mapping using Predictive Analytics

Table of Contents

1. Introduction

2. Different Predictive Analytical Methods

2.1 HotSpot

2.2 Exponential Smoothing

2.3 Linear Regression

2.4 Table Comparison between 3 Analytical Methods

3. Case Study

3.1 Predictive Analytics in Action

3.2 Methodology

4. Conclusion

5. References

1.    Introduction

With the rise of technology, we could now say that the world is getting smarter, having all information needed available to everyone which was facilitated by digitization which resulted to a diverse and massive datasets called the big data. This has been widely talked about in academic journals and industry articles, but not so common when we associate it with criminology due to several factors such as the law, ethics, privacy, etc. This is the reason why we decided to report on how the police force can benefit in using their big data with predictive analytics to make them proactive instead of becoming reactive, specifically in predicting where crimes might occur. Making use of predictive analytics can allow all emergency response units to respond intelligently and possibly in the shortest response time possible. (Kotevska et al., 2017; Groff & La Vigne, 2002; Berk & Bleich, 2012; Brayne, 2017)

Data are now being utilized more for predicting rather than for explaining in order to get more out of it. Thus, having the ability to forecast, even as simple as where crimes would occur, provides better strategical and tactical perspective for law enforcements. Mapping these crimes would enable them to deploy officers accordingly, enabling officers to patrol more effectively as they will have the knowledge to identify what to look out for. Moreover, forecasting or predicting, even with the simplest approach, has proven to be better than personal judgement since these predictions are supported by historical data and different algorithms. (Kotevska et al., 2017; Groff & La Vigne, 2002; Berk & Bleich, 2012; Brayne, 2017; Malik et al., 2014)

There are predictive analytical methods or approaches that can be used to identify areas where crimes would occur. But, for this paper, we will be exploring on 3 methods which are: Hot Spots, Exponential Smoothing, and Linear Regression. We will then identify which method would best provide law enforcements a better knowledge on where a crime would possibly occur, further explaining why the chosen method is best compared to others. The paper will end with elaborating on the importance of predictive analytics to the criminology department in preventing occurrence of future crimes.

2.    Different Predictive Analytical Methods

Chainey and Ratcliffe (2013) emphasized that a progressive blend of practical criminal justice issues with the research field of geographical information systems (GIS) and science can be known as crime mapping. Several methods which can be done the crime mapping using the predictive analytics modules have been identified. In this section, few of those identified methods will be introduced.

2.1 HotSpot

The easiest method of forecasting for the special task forces and the police forces is assumed to be the hot spots of previous day as the hot spot of the next day. (Groff & Vigne, 2002) Most of the time special task forces are being deployed to the places that the crime analysts show, using the hot spot maps. To gain the most accurate and less error hot spots to predict the crime spots, the analysts should be able to use large amount of data which gives the meaning that the data which comprises only about a weak or a month will not be sufficient to predict accurately. (Groff & Vigne, 2002) Hence, it is important to have data worth at least a year for crime prediction. For this prediction, spatiotemporal event data such as crime reports should be used. (Groff & Vigne, 2002; Maciejewski et al., 2010) In these kind of data, events consists of locations linked with the time along with / or not the space. Data are usually processed by the crime analysts based on the spatial location which can the suburb, state or the country, if not it can be processed based on the time such as date, week, or the year.

The most common method or the framework used by the crime analysts for the purpose of identifying hotspots is the use of GIS (Geographical Information Systems). In GIS you can create circles, and if one particular area shows a higher rate of crime, the radius of the circle will be higher than the rest of the circles. (Maciejewski et al., 2010) The only drawback that we can see from this method is that there is a chance that the circles may overlap and make it impossible to separate the hotspots. Recently it has been recorded that the raster GIS method has shown a rapid increase in visualising the hotspots for the crime analysts. Using this method, after the event distribution; 3D array could be used for further predictions. (Maciejewski et al., 2010) That allows the crime analysts to find out whether the hotspots can be persistent.

However, it has been identified that the hotspot view is the basic or the default view (Maciejewski et al., 2010) that the crime analysts use to predict the locations or the dates to deploy the necessary task forces to the needful places; based on the hypothesis the crime analysts generate saying that the areas in highlighted colour (often in red colour) must be having a higher risk for a particular kind of risk.

To be a crime predictive analyst it is essential to have basic GIS skills and knowledge about the softer or raster GIS (mapping). The most important advantage of this method can be shown as the ease of computation and the interpretation.

2.2 Exponential Smoothing

Exponential smoothing is a univariate method which means that only uses the previous data from a single variable (time series data) to predict the future events. (Maciejewski et al., 2010) Crime analysts believe that this is by far the easiest method because it is straightforward yet the least accurate method as well since it uses only one variable for the prediction. In 1950s Brown introduced the simplest exponential smoothing algorithm which is known as the single exponential smoothing. (Bermúdez et al., 2006) The extended version of the single exponential smoothing method which includes the time trend as well, was introduced by Holt which is known as the Holt’s Linear method. Holt-Winter double exponential smoothing method was introduced as the extended version of Holt’s linear method. (Bermúdez et al., 2006) This framework allows easy calculation and model selection criteria such as AIC which is known as Akaike’s Information Criteria and BIC which is known as Bayesian Information Criterion. (Brownlee, 2018)

Exponential smoothing method can be used to analyse a small areal unit such as a precinct. (Maciejewski et al., 2010) Crime data analysts require area level data to predict the future crime locations where they can deploy the necessary special task forces to minimise the crime in those specific areas. To use this method; they need software which has the ability of statistical programming and geographical information system to aggregate the available data and display the results. (Maciejewski et al., 2010) It is essential that the crime analysts who deploy or use this exponential smoothing method has the knowledge of forecasting proficiency along with the GIS knowledge. Using exponential smoothing it enables statistical analyses to consider changes over this. Even though this method is not accurate and not suitable for large amount of data; this method can still be used to analyse small changes in crime.

2.3 Linear Regression

The basic definition of linear regression is that it is a type of statistical analysis that is often used in predictive analytics where it creates a predictive model of a data showing the trends in it. It shows the relationship between a dependent variable and its independent variables.  This tool focuses on forecasting a trend, an effect, and identifying the strength of predictors. Furthermore, results can be visualised through scatter plot, box plot, and density plot. This tool is useful when it comes to forecasting or predicting crime patterns which law enforcers could prepare for because: (1) in forecasting, it already incorporates possible sudden changes in crime rates, and (2) it predicts changes that are very different from the normal variation. (Statistics Solutions, 2013; Prabhakaran, 2016;  Techopedia; Shingleton, 2012)

In using this though, the data to be used should be grid level and the area of this grid cell should be as small as 4000 by 4000 foot grid cells. The person tasked to do this should at least be an intermediate level in statistics and desktop GIS. Software requirements are also necessary to perform this method such as Spatial Statistical Program which can compute spatial lag, and desktop GIS to aggregate results and display data. (Groff & Vigne, 2002)

2.4 Table Comparison between 3 Analytical Methods

3.    Case Study

3.1 Predictive Analytics in Action

While looking for a data to work on to show how a simple analytical visualization could aid in predicting future outcomes, and to test the theories, we came across the crime data of Australia through Australian Bureau of Statistics and Crime Statistics. However, their downloadable files were heavily formatted which we could not use as our dataset. Therefore we had to expand our search to find a dataset that would fit the needs of the software that we are using which is SAP Analytics Cloud. This tools is perfect to demonstrate what we could achieve with predictive analytics with minimal coding and quick results. Eventually, we found data from Analyze Boston, which has over 140 datasets. They have crime dataset dating from 2015 to present which could provide us with good analytical results.

Once we downloaded the data, we had over 330,000 rows of data, after cleansing the dataset (removing unnecessary and empty fields), we were left with a little bit over 300,000 rows of data. Moreover, this dataset is great because it included longitude and latitude coordinates of where specific crime occurred which is important for mapping purposes. (See Figure 1)

Our aim was to find out the places/locations where the crime (i.e. murder, theft or an assault) has a higher risk to happen, so the crime departments such as police and special task forces can pay more attention to the identified area.

In the literature review, we identified different methods that we can employ to predict the locations which has the higher probability of crime occurrence. And we believe, to carry out this research; most suitable method would be hot spot method. We came to this decision after considering the ease of computation and interpretation and the understanding that it produces accurate information for the decision making process. The crime analysts do not have to be highly skilled or experienced professionals in order to use this method, which made it easier for us to carry out the research and predict the hot spots of the higher crime rate. The basic skills that we should possess are the geographical information systems skills and having knowledge about raster GIS mapping with spatial statistics. Apart from these benefits; we identified that the hot spot approach is a much efficient method than the other two approaches that we introduced in the paper. (See Section 2)

3.2 Methodology

For the process of doing the hot spot approach, we had to do modification to the data first in SAP Analytics Cloud. Despite having a location dimension already in place, we had to do a Geo Enrichment transformation by coordinates so we could access the correct location dimension that the system will be able to recognize and use the latitude and longitude accordingly. (See Figure 2)

This will then provide us with a new system recognized location dimension with the type ST-Point which is important for geo mapping. (See Figure 3)

We then proceeded to creating a geo map, the goal was to identify which area in Boston had the most crime incidents recorded. (See Figure 4)

With just a few steps and using geo map, we are able to identify which areas are heavily involved with crime. With this type of visualisation, police officers can quickly identify which places need more surveillance and protection.

Moreover, they could de-cluster this results into individual crime. (See Figure 5)

4.    Conclusion

In this research report it is identified that crime mapping is an essential visualization that the governments and the police forces should look into. Furthermore, our aim for this research paper was to introduce an easy method for the police forces to predict the locations where there is a higher risk for crime occurrence. To achieve this purpose, as mentioned in the introduction as well; we referred to other scholarly articles and referenced the most relevant three predictive analytics methods which are hot spot, exponential smoothing and linear regression.

We were able to identify the advantages of each method that we introduced along with the software requirements and the skills or the experiences that the crime analyst who is going to predict using those methods should possess. During this study, we were able to choose “hot spot” as the best method to carry out the small-scale research, mainly due to its higher accuracy and efficiency when compared with the exponential smoothing and linear regression, and the fact that it can be analysed quickly. The “hot spot” method allows the users to predict using the GIS (Geographical Information System) and spatial statistical software.

After the identification of the most suitable method for the research; the report describes the methodology that we followed in order to get the desired results from the data set we retrieved. For the purpose of this research report, SAP Analytics Cloud was used as the tool to identify and interpret the hot spots.

Furthermore, one of the obstacle that we had to face with this research report was the lack of journal articles that are up to date in relation to criminology and predictive analytical methods. For future direction, this paper could aid in simple analysation or prediction for Australian crime.

5.    References

• Analyze Boston, Crime Incident Reports (August 2015 – To Date) (Source: New System), viewed on 7 October, <https://data.boston.gov/dataset/crime-incident-reports-august-2015-to-date-source-new-system>
• Berk, R & Bleich, J 2012, ‘Forecasts of Violence to Inform Sentencing Decisions’, University of Pennsylvania.
• Bermudez, JD, Segura, JV, & Vercher, E 2006, ‘Improving Demand Forecasting Accuracy Using Non-Linear Programming Software’, Journal of the Operational Research Society, vol. 57, pp. 94-100.
• Brayne, S 2017, ‘Big Data Surveillance: The Case of Policing’, American Sociological Review, vol. 82, no. 5, pp. 977-1008.
• Brownlee, J 2018, ‘A Gentle Introduction to Exponential Smoothing for Time Series Forecasting in Python’, Machine Learning Mastery, weblog post, viewed 8 October, <https://machinelearningmastery.com/exponential-smoothing-for-time-series-forecasting-in-python/>
• Chainey, S & Ratcliffe, J 2013, GIS and Crime Mapping, John Wiley & Sons Ltd, England.
• Dzemydiene, D, & Rudzkiene, V 2002, ‘Multiple Regression Analysis in Crime Pattern Warehouse for Decision Support’, Lecture Notes in Computer Science Database and Expert Systems Applications, pp. 249–258.
• Groff, E & La Vigne, N 2002, ‘Forecasting the Future of Predictive Crime Mapping’, Crime Prevention Studies, vol. 13, pp. 29-57.
• Kotevska, O, Kusne, G, Samarov, D, Lbath, A, & Battou, A 2017, ‘Dynamic Network Model for Smart City Data-Loss Resilience Case Study: City-to-City Network for Crime Analytics’, IEEE Access Special Section on Advanced Data Analytics for Large-scale Complex Data Environments, vol. 5, pp. 20524-20535
• Maciejewski, R, Hafen, R, Rudolph, S,  Larew, SG, Mitchell, MA, Cleveland, WS, Ebert, D 2010, ‘Forecasting Hotspots—A Predictive Analytics Approach’, IEEE Transactions on Visualization and Computer Graphics, vol.17, no. 4, pp. 440-453.
• Malik, A, Maciejewski, R, Towers, S, McCullough, S, & Ebert, D 2014, ‘Proactive Spatiotemporal Resource Allocation and Predictive Visual Analytics for Community Policing and Law Enforcement’, IEEE Transactions on Visualization and Computer Graphics, vol. 20, no. 12, pp. 1863-1872.
• Prabharakan, S 2017, ‘Linear Regression’, weblog post, viewed 8 October, <http://r-statistics.co/Linear-Regression.html>
• SAP Analytics Cloud, online software, used 7 October, <https://www.sapanalytics.cloud/>
• Shingleton, J 2012, Crime Trend Prediction Using Regression Models for Salinas, California, Postgraduate Thesis, Naval Post Graduate School, Monterey, California, viewed on 8 October, <http://www.dtic.mil/dtic/tr/fulltext/u2/a563653.pdf>
• Statistics Solution 2013, ‘What is Linear Regression’, weblog post, viewed 8 October, <http://www.statisticssolutions.com/what-is-linear-regression/>
• Techopedia, Linear Regression, weblog post, viewed 8 October, <https://www.techopedia.com/definition/32060/linear-regression>