See discussions, stats, and author profiles for this publication at: [Link]

net/publication/320650913

An overview on crime prediction methods

Conference Paper · May 2017

DOI: 10.1109/ICT-ISPC.2017.8075335

CITATIONS READS

65 5,134

3 authors, including:

Nor Azizah Ali Razana Alwee

University of Technology Malaysia University of Technology Malaysia
23 PUBLICATIONS 219 CITATIONS 40 PUBLICATIONS 236 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Nor Azizah Ali on 18 March 2019.

The user has requested enhancement of the downloaded file.

 An Overview on Crime Prediction Methods
Nurul Hazwani Mohd Shamsuddin1, Nor Azizah Ali2, Razana Alwee3
1,2,3
Faculty of Computing,
Universiti Teknologi Malaysia,
Johor, Malaysia.
[Link]@gmail.com1, nzah@utm.my2, razana@utm.my3

Abstract— In the recent past, crime analyses are required to tasks of discovering and detection of crimes and their relation
reveal the complexities in the crime dataset. This process will help with criminals.
the parties that involve in law enforcement in arresting offenders
However, a considerable challenge faced by law
and directing the crime prevention strategies. The ability to predict
the future crimes based on the location, pattern and time can serve enforcements is to analyze the increasing numbers of crime
as a valuable source of knowledge for them either from strategic or data accurately and efficiently [10]. The ability to analyze
tactical perspectives. Nevertheless, to predict future crime amount of crime data without using computational support
accurately with a better performance, it is a challenging task will put strain on human because human mechanism is
because of the increasing numbers of crime in present days. incapable of comprehending with millions of data [11]. Since
Therefore, crime prediction method is important to identify the the high dimensional of crime data, traditional crime analysis
future crime and reduces the numbers of crime. Currently, some
researchers have been conducted a study to predict crime based on
methods cannot be applied for identifying the future crimes.
particular inputs. The performance of prediction models can be The powerful system in predicting crime is required when the
evaluated using a variety of different prediction methods such as more data and the complex queries are being processed [10,
support vector machine, multivariate time series and artificial 11]. Therefore, developing a systematic crime analysis tool to
neural network. However, there are still some limitations on their identify the crime patterns effectively and promptly for future
findings to provide an accurate prediction for the location of crime patterns detection is needed. This includes the capability
crimes. A large number of research papers on this topic have
of the system to analyze or evaluate a large amount of data
already been published previously. Thus, in this paper, we
thoroughly review each of them and summarized the outcomes. Our and information from variety sources.
objective is to identify current implementations of crime prediction Hence, this study was conducted to review and examine
method and the possibility to enhance it for future needs. the current methods used in crime prediction analysis. The
organizations of this paper are as follows. Section I describes
Keywords—crime analysis; crime prediction; prediction the introduction of the study. Section II discusses the overview
methods
of crime prediction methods. In Section III, we thoroughly
review the previous study related to crime prediction and
I. INTRODUCTION summarized the important outcomes. The discussions of this
Crime is one of the major issues is continuing to grow in paper are written in Section IV and the conclusions are given
intensity and complexity [1]. In the recent years, crime is one in Section V.
of the social problems influencing the nature of life and
economic development in a community [1]. Crime can be II. CRIME PREDICTION METHODS
divided into a few types such as crime against properties
A. Support Vector Machine (SVM)
(theft, burglary, and robbery) and crime of aggression
(homicides, assaults and rape) [2]. The availability of Recently, support vector machine (SVM) is a nonlinear
model and it has been performed well for time series
information technologies has enabled law enforcement to
prediction, classification tasks and regression [12]. The
collect detailed information of crime data [3]. With the performance of the SVM is compared with the performance of
increasing numbers of crimes nowadays, crime analysis is both recurrent neural network model (RNN) and
needed which comprises measure and procedure that intend to autoregressive moving average (ARMA) in time series
reduce the risk of crime [4]. Crime analysis can be done prediction [13]. ARMA model is a linear behavior but it is
through both quantitative and qualitative methods [5, 6]. easy and fast in use while RNN can be in principle model
Qualitative approaches in predicting crime such as scenario linear but they are usually difficult to train and does not lead
writing or environmental scanning are valuable in identifying to one unique or global solution due to differences in their
the future of criminal activity [7]. Meanwhile, quantitative initial weight set [13]. Besides that, SVM has been performed
method is used to predict the crime rates in future specifically. well in prediction of time series because they can model
nonlinear relations in a stable and efficient way [14].
Moreover, crime analysis is a practical approach to analyze
Furthermore, the SVM is educated as a convex optimization
and identify the pattern of crimes [8]. The study by [9] stated problem resulting in a global solution which in many cases
that crime analysis is part of crime prevention which has the

978-1-5090-6255-3/17/$31.00 ©2017 IEEE

output unique solutions [15]. Thus, SVM was applied in III. RELATED WORKS
predicting of crime hot-spots [16] and predict the common Many different methods for crime prediction have been
diseases (diabetes and pre-diabetes) [17].
proposed by researchers. However, there are still has some
limitations in these methods.
B. Fuzzy Theory
According to [16], Support Vector Machine method can be
Furthermore, to build the prediction system using fuzzy used in hot-spots prediction to predefine a level of crime rate
modeling mentioned in [18] requires numerical inputs. The IF- and given the percentage of data. A subset of the crime
THEN regulations formed have vague predicates in their datasets are selected (percentage or number of data crimes)
antecedent part and a linear or quadratic combination of the and classify each of selected data point based on the
antecedent variables for consequent part [19]. There is no need predefined level of crime rate. The data point which has above
to defuzzify the output for the consequent parts since their
than predefined rate are positive or classified as hotspot class
rules are crisp value rather than vague and fuzzy [19].
Meanwhile, fuzzy modeling requires less computational and the data point which have below than predefined rate are
efforts and excellent learning capabilities. It provides similar negative or non-hotspot class. The parameter selected will be
degree of robustness and accuracy together with lesser used as the training in SVM classification. As a result, linear,
computational complexity as compared with various other polynomial and gaussian kernel function has been chosen to
methods [20]. The prediction system can be affected by compare the performance of one-class SVM for predicting the
increasing the number of inputs [21]. The characteristic of the hotspot crime of location. The datasets are from the Internet
fuzzy modeling is used to improve the prediction efficiency. public datasets (2 datasets, Columbus and ST. Louis).
Thus, fuzzy was applied for forecasting crime [22], predict the However, using this method is still slow and computationally
quality of software [23] and predict the price of stock [24]. expensive even though it has been modified.
Fuzzy time series was applied by [22], in orders to
C. Artificial Neural Network (ANN) discover a crime pattern in community. To reduce the
According to [25], artificial neural network (ANN) is a computational overhead in this method, the simplified process
model inspired by biological neuron and act as the human and model that include simple arithmetic operations has been
brain in decision making. It consists of a large number of used. This method utilized seventeen years historic data of
processing components that work together to solve problems crime incidents (cases of murder in Delhi City). Hence, the
[25]. Meanwhile, ANN model is based on the prediction by parameters used are actual registered cases and years. The
smartly analyzing the trend from an already existing three different sets of data obtained by partitioning it into five
voluminous historical set of data [26]. ANN has more flexible intervals (Scheme-I), ten intervals (Scheme-II) and twenty
and general functional forms can effectively deal with than the intervals (Scheme-III) implemented. The prediction of results
traditional statistical methods [27]. Frequently, traditional obtained from Scheme-II and Scheme-III are quite acceptable
statistical prediction models have limitations functions to while the result of scheme-I incline to slightly over forecast
estimate the relationship between the inputs (the past values of with an average absolute error (-0.323). However, this method
the time series or other relevant variables) and the outputs (the works on binary transaction data only for example 0 or 1.
future values) due to the complexity of the real system but Prediction crime using artificial neural network (ANN) as
ANN can be used as an alternative method to resolve these one of proposed method mentioned in [30]. This method
limitation functions [28]. Thus, ANN was applied in introduced by focusing on geographical areas that outperform
predicting of groundwater level [29], predict the geo-temporal traditional policing boundaries for crime prediction.
variations of crime and disorder [30] and predict osteoporosis Therefore, ANN can be educated using geographical clusters
in the elderly [31]. of crime data to ease predictive modeling. Hence, the scanning
algorithm based on geographical crime incidents used to
D. Multivariate Time Series identify clusters with relatively high level of crime hotspots.
The study conducted by [32] mentioned that multivariate Generally, ANN reveals a capacity to model the trends with
time series is one of the statistical tools to study the behavior each cluster. In this study, the dataset used are 18,498 violent
of time-dependent data and predict the future values based on incidents (criminal damages, violence against the person)
history of variations in the data. According to [33], included a number of variables related to time, day, month,
multivariate time series can be significantly used better than location and weather. The result shows that the comparison of
using univariate time series in prediction. This method is not mean standard error (MSE) between ANN (9.94) and random
essential for recovering dynamics but only one time series walk (22.50). Thus, ANN predicts accurately more than
should be needed according to the embedding theorem. random walk. Nevertheless, this method takes a long time in
Multivariate time series data are available in many practical training phases.
situations such as economic data and physiological data are For analyzing a large number of crime data in different
usually multi-dimensional [34]. Thus, this method was applied time points, an unsupervised method by using multivariate
in predicting of model for cash-flow data, testing the statistical time series is introduced by [32]. This method usually is based
significance of multivariate time series analysis technique for on parametric Minkowski model and dynamic time wrapping
epileptic seizure prediction and predicts the business failures (DTW). It has been proposed to find similar crime trends
[34]. among various crimes cases and used the obtained information
for future crime trends prediction. The effectiveness of this this approach is difficult to identify the probability of burglary
method has been proved using Indian crime dataset provided activity and specific locations.
by the Indian National Crime Records Bureau (29 districts).
The parameter of this model is the different of dimensions IV. DISCUSSION
data weights. Their results showed the comparison between Solving a crime problem is a complex task that requires
DTW with Euclidean (equal weight age) and DWT with human experience and intelligence and also methods that can
parametric Minkowski model (different weight age). This help them with crime detection problems [38]. Besides that,
method proposed for finding similar crime trends and predict the prediction of future crime patterns or trends involves the
crime trends efficiently where the dimensions of data do not changes of crime rate from one year to another year and used
have equal weight age. However, this method is difficult to the prediction methods to find out those changes into the
handle a missing value in order to get more accurate results. future [39]. As discussed in [40], the current problem is to
In addition, [35] introduced Bayesian Network based on identify methods that can efficiently and accurately analyze
bayesian learning theory as one of crime prediction model. this growing volume of crime dataset. The proposed study by
This method can be used to build effective mathematical [41] proved that prediction crime using neural network
models to understand the behaviors of serial crimes. There are method is generally high in accuracy and fast in evaluation of
many potential factors affecting the selective of criminals for the learned target function.
the future crime location. The model has been testing using The dataset for crime prediction which has been used is
serial crime dataset found in Gansu, China. Hence, the usually found in official records. There is an important part for
parameters used are characteristics of the victims (age, gender, experimentation to develop a more scientific and exact
jobs and race) and characteristics of crime areas (residence, procedure for obtaining information in future [42-44]. The
school, bus stop, hotel and hospital). Furthermore, the given study conducted by [12], to get a satisfactory model for crime
predicted areas (yellow, green and red color) will help police prediction analysis and to get accurate results of analyzing
to arrest criminals but it is depends on geographical factors crime dataset, it requires huge historical data that can be used
selected. Nevertheless, this method is totally depending on for training and testing the model. Hence, [25] mentioned that
selection of parameter. Therefore, it is may lead to some ANN can provide a better accuracy because of larger volumes
deviation during conducted the experiment so the more factors in data crimes.
unrelated to the geography should be considered to improve According to [19], the accuracy of crime prediction is
the accuracy of the model. depending on dataset. If dataset used is a huge number, the
As discussed in [36], decision tree is one of the supervised model training is achieved the high level in accuracy while if
methods used to handle the complex classification and the dataset used is a small number, the low level of model
produce reasonable classification tree. It provides the tree training is achieved. Moreover, the accuracy of prediction is
structure of classification and divides a dataset into smaller depending on the level of training data based on their amount
subsets. The decision tree helps police to discover the crime or dimension data. The accuracy will be more accurate if the
pattern and predict the future trends. The final result model is highly trained while if the model is not highly
represented as a tree with decision nodes and leaf nodes which trained, the result of accuracy will be low [44].
decision nodes have two or more branches and leaf nodes For future works involving crime prediction algorithm in
represent a decision or classification. The dataset used in their crime analysis field, it is hoped that researchers can propose
testing and training phases were extracted from the Internet. more hybrid methods between crime prediction algorithm in
The parameters used in dataset are heartbeat rate, facial order to get best finding data and results. As mentioned by
emotion, state and voice tone. The result of crime cases has [30], to produce better result, the hybrid methods are highly
been classified into two classes such as neutral and danger. recommended compared to the filter methods.
However, this method does not work well on all type of
datasets. V. CONCLUSION
The study by [37] applied a logistic regression model to The crime analysis is sensitive domain where efficient for
examine the relationship between the predicting factors of prediction and classification to analyze the increasing numbers
crimes and burglary occurrence probability. The factors used of crime data. Hence, the crime prediction methods will be
including time of day (category data), day of the week evaluated and analyzed by the systematic tool in crime
(discrete), barriers (physical structures that will interfere an analysis. The biggest challenge facing by many law
individual’s egress from a targeted residence), connectors (the enforcement is how to efficiently and accurately analyzing the
amount of access bridges, streets and pathways relative to the increasing volumes of crime data. This research work focuses
targeted residence) and repeat victimization (the occurrence of on reviewing a crime prediction analysis tool for many
the offense at the same residence following the initial offense scenarios using particulars crime prediction methods that can
over the calendar year, same category data). The datasets for help law enforcement to efficiently handle crime incidents.
testing were collected in burglary incidents in 2010 from a Therefore, a crime analysis should be able to identify the
local police report of Florida City. The model has been shown crime patterns as fast as possible and in an effective manner
the various degrees of significance in terms of predicting the for future crime detection. Using a wide range of methods
occurrence within different ranges. However, the limitation of such as artificial neural network, fuzzy or support vector
machine, it is possible to find helpful information to help in [17] W. Yu, T. Liu, R. Valdez, M. Gwinn and M.J. Khoury, “Application of
support vector machine modeling for prediction of common disease: the
crime prediction problems. To our knowledge, there is no case of diabetes and pre-diabetes,” in BMC Medical Informatics and
standard method that can solve the problems in different Decision Making, Vol. 10, No. 16, 2010.
datasets of crime. In this paper, we report on result of an [18] S.M. Chen and C.C. Hsu, “A new method to forecasting enrollments
extensive comparison of crime prediction methods. using fuzzy time series,” in International Journal of Applied Science and
Furthermore, for future research, we propose to expand this Engineering, Vol. 2, No. 3, 2004, pp. 234-244.
study to enhance the crime prediction methods in order to [19] P. Saxena, K. Sharma and S. Easo, “Forecasting Enrollments based on
Fuzzy Time Series with Higher Forecast Accuracy Rate,” in
resolve the limitation of current methods to obtain more International Journal of Computer Technology and Applications, Vol. 3,
accurate result and good performance. No. 3, 2012.
[20] C. Yu-Chuan and C. Shyi-Ming, “Temperature prediction based in fuzzy
ACKNOWLEDGMENT clustering and fuzzy rules interpolation techniques,” in International
Conference on Systems, Man and Cybernatics, pp. 3444-3449, 2009.
This material is based upon work supported by Grant
[21] Y. Luo and Shibo, Z, “The fuzzy regression prediction of the city road
University Teknologi Malaysia (UTM) (Ref No: traffic accident,” International Conference on Industrial Mechatronics
R.J130000.7828.4F509. and Automation, pp. 121-124, 2009.
[22] A.K. Shrivastav, “Applicability of Soft Computing Technique for Crime
Forecasting: A Preliminary Investigation,” in International Journal of
REFERENCES Computer Science & Engineering Technology, vol. 9, no. 9, 2012, pp
415-421.
[1] A. Bogomolov, B. Lepri, J. Staiano, N. Oliver, F. Pianesi and A.
Pentland, “Once upon a crime: towards crime prediction from [23] X. Yuan, T.M. Khoshgoftaar, E.B. Allen and K. Ganesan, “An
demographics and mobile data,” in Proceedings of the 16th international application of fuzzy clustering to software quality prediction,” in IEEE
conference on multimodal interaction, 2014, pp. 427-434. Symposium on Application-Specific Systems and Software Engineering
Technology, pp. 85-90, 2000.
[2] S. Sathyadevan, M.S. Devan and G.S. Surya, “Crime analysis and
prediction using data mining,” in International Conference on Networks [24] H.N. Nhu, S. Nitsuwat and M. Sodanil, “Prediction of stock price using
& Soft Computing, 2014, pp. 406-412. an adaptive Neuro-Fuzzy Inference System trained by Firefly
[3] P. Thongtae and S, Srisuk, S. “An analysis of data mining applications Algorithm,” in International Conference in Computer Science and
Engineering, 2013, pp. 302-307.
in crime domain,” in International Conference on Computer and
Information Technology Workshops, 2008, pp. 122-126. [25] A. Kumar and A. Narain, “Designing of Controller based on Artificial
[4] V. Grover, R. Adderley and M. Bramer, “Review of current crime Neural Network for liquid Level System,” in International Journal of
prediction techniques,” in Applications and Innovations in Intelligent Artificial Intelligence and Neural Network, Vol. 3, No. 3, 2013, pp. 17-
Systems, Vol. 14, 2007, pp. 233-237. 22.
[5] H. Liu, D.E. Brown, “Criminal incident prediction using a point-pattern- [26] J. Caulkins, J. Cohen, W. Gorr and J. Wei, “Predicting Criminal
based density model,” in International Journal of Forecasting, Vol. 19, Recidivism: A Comparison of Neural Network Models with Statistical
2003, pp. 603-622. Methods,” in Journal of Criminal Justice, Vol. 24, No. 3, 1996, pp. 227-
240.
[6] G.B. Vold, “Prediction Methods Applied to Problems of Classification
[27] F. Mingjian, Z. Guocheng, Z. Xuxu and Y. Zhongyi, “Study on Air Fine
Within Institutions,” in Journal of Criminal Law and Criminology, Vol.
26, 1951, pp. 202-209. Particles Pollution Prediction of Main Traffic Route using Artifical
Neural Network,” in International Conference on Computer Distributed
[7] A. Babakura, M.N. Sulaiman and M.A. Yusuf, “Improved method of Control and Intelligent Environmental, 2011, pp. 1346-1349.
classification algorithms for crime prediction,” in International
Symposium on Biometrics and Security Technologies (ISBAST), 2014, [28] X. Hong and R. Qunhua, “Flood Level Prediction on the Basic of the
pp. 250-255. Artificial Neural Network,” in International Conference on Information
Science and Engineering, 2009, pp. 4887-4890.
[8] H. Chen, W. Chung and J.J. Xu, "Crime data mining: a general
framework and some examples." Computer 37.4, 2004, pp. 50-56. [29] A. Kumar, R. Ranjan and S.A. Kumar, “A rainfall prediction model
using artificial neural network,” in IEEE Control and System Graduate
[9] F. Barreras, C. DÕaz, A. Riascos and M. Ribero, “Comparison of Research Colloqium, 2012, pp. 82-87.
different crime prediction models in Bogotá,” 2016.
[30] M. Chitsazan and G. Rahmani, “Groundwater level simulation using
[10] N. Malleson, A. Heppenstall and L. See, “Crime reduction through artificial neural network: a case study from Aghili plain, urban area of
simulation: An agent-based model of burglary,” Computers, Gotvand, south west Iran,” in Journal of Geopersia, Vol. 3, No. 1, 2013,
Environmental and Urban Systems, Vol. 34, No. 3, 2010, pp. 236-250. pp. 35-46.
[11] J.L. Nasar and B. Fisher, “Hotspots of fear and crime: A multi-method [31] J.S. Chiu, Y.C Li and Y.F. Wang, “Applying an artificial neural network
investigation,” in Journal of Environmental Psychology, Vol. 13, No. 3, to predict osteoporosis in the elderly,” in Studies in Health Technology
1993, pp. 187-206. and Informatics, Vol. 124, 2016, pp. 609-614.
[12] P. Wang, R. Mathieu and H.J. Cai, “Predicting Criminal Recidivism [32] B. Chandra, M. Gupta and M.P. Gupta, “A Multivariate Time Series
With Support Vector Machine,” in International Conference on In Clustering Approach for Crime Trends Prediction,” in International
management and Service Science, 2010, pp. 1-9. Conference on In Systems, Man and Cybernatics, 2008, pp. 892-896.
[13] Q. Li-Na, “Software reliability prediction model based on PSO and [33] D.F. Greenberg, “Time series analysis of crime rates,” in Journal of
SVM,” in International Conference on Consumer Electronics, Quantitative Criminology, 17(4), 2001, pp. 291-327.
Communications and Networks, 2011, pp. 5236-5239.
[34] H. Corman and H.N. Mocan, “A time-series analysis of crime,
[14] K. Zhu and J. Zhang, “Predicting the potential locations of the next deterrence, and drug abuse in New York City,” in The American
crime based on data mining,” in International Journal of Digital Content economic review, 90(3), 2000, pp. 584-604.
Technology and Its Application, Vol. 6, No. 20, 2012, pp. 574-581.
[35] R. Liao, X. Wang, L. Li and Z. Qinh, “A Novel Serial Crime Prediction
[15] W. Weiwei, “Time Series Prediction Based on SVM and GA,” in Model Based on Bayesian Learning Theory,” in International
International Conference on Electronic Measurement and Instruments, Conference on Machine Learning and Cybernatics, 2013, pp. 1757-
2007, pp. 2307-2310. 1762.
[16] K. Kianmehr and R. Alhajj, “Crime Hot-spots prediction using support [36] A. Nasridinov, S.Y. Ihm and Y.H. Park, “A Decision Tree-Based
vector machine,” in IEEE International Conference on Computer Classification Model for Crime Prediction,” Information Technology
Systems and Applications, 2006, pp. 952-959.
 Convergence, Lecturer Notes in Electrical Engineering, vol. 253, 2015, [40] A. Malathi, S.S. Baboo, “An Enhanced Algorithm to Predict a Future
pp. 531-538. Crime using Data Mining,” in International Journal of Computer
[37] D. Antolos, D. Liu, A. Ludu and D. Vincenzi, “Burglary Crime Analysis Applications, Vol. 21, 2011, pp. 1-6.
Using Logistic Regression,” in Human Interface and the Management of [41] C.H. Yu, “Crime Forecasting Using Data Mining Techniques,” in
Information. Information and Interaction for Learning, Culture, International Conference on Data Mining Workshop, 2011, pp. 779-786.
Collaboration and Business, Lecture Notes in Computer Science, vol. [42] E.R. Groff and N.G. La Vigne, “Forecasting the future of predictive
8018, 2013, pp. 549-558. crime mapping. Crime Prevention Studies, Vol. 13, 2002, pp. 29-58.
[38] K.B.S. Al-Janabi, “A Proposed Framework for Analyzing Crime Data [43] K.C. Baumgartner, S. Ferrari and C.G. Salfati, “Bayesian Network
Set using Decision Tree and Simple K-Means Mining Algorithm,” in Modeling of Offender Behavior for Criminal Profiling,” in International
Journal of Kufa for Mathematics and Computer, Vol. 1, No. 3, 2011, pp. Conference on Decision and Control and the European Control, 2005,
8-24. pp. 2702-2709.
[39] N.M.M. Noor, R. Astari, M. Norizan, M.S.M Yazid and A.M. Lazim, [44] R. Iqbal, M.A.A. Murad, A. Mustapha, P.H.S Panahy, N.
“A framework of decision support for crime forecasting in Malaysia,” in Khanahmadliravi, “An Experimental Study of Classification Algorithms
Proceedings of the 2nd European Conference for the Applied for Crime Prediction,” in Indian Journal of Science and Technology,
Mathematics and Informatics, 2011, pp. 31-35. Vol. 6, No. 3, 2013, pp. 4219-4225.

View publication stats