1

Machine Learning Approaches

For
Detect Crime Patterns
N.D. Waduge, Dr. L. Ranathunga

Abstract: Crimes are one of the major threat to the society and also for the civilization. The traditional crime solving
techniques unable to live up to the requirement of existing crime scenario. Most challenging area in this crimes is
identifying the sets of crimes committed by the same individual or same group. Criminal are humans. Then they tend to
do the same work in the same way. Surveys states that 50% of the crimes done by the 10% of criminals. In this scenario
Machine Learning can be used to identify the patterns of crimes. The data to feed this Machine Learning approach can
be taken from past crime records, social media sentiment analysis, weather data etc. There are five steps in crime
prediction by using Machine Learning. Those are data collection, data classification, pattern identification, prediction
and visualization. The limited resources in law enforcement authorities can be used effectively by using crime prediction
methods. Especially data collected through the social media shows that prediction of a crime can be done using Machine
Learning and it shows reasonable percentage of hits.

Index Terms— Modus Operandi, Machine Learning, Supervised Learning

same locales and high crime risk areas repeatedly. This is

called “Modus Operandi” (MO). Series of crime have some
I. INTRODUCTION1 common attributes which can be used to characterize the
modus operandi of the criminal.
A ccording to the crime analysis there are four major
types of crimes. They are murder, forcible rape,
robbery and aggravated assault. Apart from these
Therefore this paper attempts to identify a machine learning
approach to predict the crime series patterns by using past
major areas there are drug offences, traffic violations, data related to the crimes. Identify the geographical
damaging to private and public properties. According to the ‘hotspots’ where a crimes can happened frequently.
time based analysis there are two types of crime. Which are
accidental crimes and planned crimes. Studies on these The rest of the paper organized as follows. Section 2
planned crimes observed that there are patterns of describes the related work. Section 3 provides a description
happening those kind of crimes on specific geographical about methodology while Section 3 gives a conclusion.
areas. Persons who intended to do crimes choose seclude
places for committing the crime where the police and other
law enforcement patrols are less. Because of this reason
there is a higher probability of predicting the crimes which II. RELATED WORK
going to happen in the future.
In Los Angeles police department officers are currently
However, earlier days the data about crimes are based on using crime prediction system called predpol[3]. Also the
the police reports, newspapers, reports and articles. Those Cambridge police department done a research called “Series
all were in hand written hard copy format, but nowadays Finder”. There are significant number of researches done
police and other law enforcement authorities are about crime prediction using “Modus Operandi” (Method of
maintaining a soft copy of these data along with the hard Operation) [2]. Some researches show that in the data
copy. Those soft data generation in these days are very high. gathering process, it can be used the social medias [4]. Most
These data are just not a bulk of data but valuable of these crime prediction methods showed considerable
information that can be used to predict upcoming crimes success rate in crime predictions.
and solve exiting. Criminals are human beings, since they
are trying to repeatedly do the same thing. They will use

1
 habits that the offender follows, and is a type of motif used
to characterize the pattern. As more crimes are added to the
III. METHODOLOGY set, the M.O. becomes more well-defined. Following 21
Data Collection characteristics have been extracted as MO table to their
research.
The data collection has been done for crime prediction by
using two major data sources. Those are
 Crime data records from law enforcement authorities
 Social media analysis
 IoT devices
The largest data set maintained by a law enforcement
authority is “Crime data 2001 to present” [5] by Chicago
Police department. There are significant number of Crime
Prediction systems tested in Chicago city. [6]
In Sri Lanka these data can be collected from Criminal
Records Office (CRO), but those are not in digital format.
Because of that reason it is mandatory to convert those
written files to digital format before use.

Other major way of collection of data to crime prediction

are social medias. Most of the researches has been used
Twitter [6] [7] [4] because of the flexibleness of the APIs.
They have used users tweets for a selected geographical
area for predict crimes. In a research from South Africa,
they have used Twitter (tweets) to collect data of crimes in Classification and Pattern Recognition
South Africa. The tweets’ data about crimes in specific
geographical areas being recorded. According to those In the above research Clustering Algorithms has been used
records they have determined that which specific to check whether it is possible to short list crimes using
geographical areas are famous for what specific crime MO. Most of the researches including the above one
pattern. implemented with ‘K-Means’ clustering [9]. K-means is a
widely known clustering partitioning approach. For
applications in which a number of clusters are required [10].
There was a one research conducted about how weather can Some of the researches used supervised learning methods,
be a reason and how weather analysis can be used to while some used un-supervised learning methods. “Apriori
forecast crimes. [7] In this research they sorted out that in Algorithm”, Decision Trees has been used in a research
hotter days there is an increment of occurring crimes. done by “Amrita Center for Cyber Security” [11] by
Following set of constraints has been checked to the supervised learning method. Apriori can be used to
relevancy for crimes. determine association rules which highlight general trends
in the database.
 Minimum, Maximum and Mean temperatures in Also some researches conducted in supervised learning
Fahrenheit methods by using “Naïve Bayes” classifier and “Back
 Humidity Propagation Algorithm”. Back Propagation Algorithm is a
 Sea level pressure model of reasoning based on the human brain [12].
 Visibility per Miles
 Wind speed in Mph
IV. CONCLUSION
 Cloud Cover
 Events(Sunny, Rain, Snow, Fog) According to the results of researches done regarding to
crime prediction using Machine Learning shows higher
In a research conducted from University of Colombo about success rate often over than 50%. Because of that evidence
this crime data prediction, they have mentioned that “Modus it can be seen that Machine Learning can be used to predict
Operandi” in other words Method of Operation can be used crimes.
to predict crimes. Criminals are creatures of habit,
repeatedly using the same locales for committing
crimes, or are attracted to certain high crime risk areas
[8]. In this research they have defined MO as, is the set of
 3

REFERENCES
[1] “Types of Crimes - Overview and Discussion.”
[Online]. Available: https://www.thoughtco.com/types-of-
crimes-3026270. [Accessed: 22-Jun-2017].
[2] M. Munasinghe, H. Perera, S. Udeshini, and R.
Weerasinghe, “Machine Learning based criminal short
listing using Modus Operandi features,” 2015, pp. 69–76.
[3] “Predict Crime | Predictive Policing Software,”
PredPol. [Online]. Available: http://www.predpol.com/.
[Accessed: 22-Jun-2017].
[4] S. Aghababaei and M. Makrehchi, “Mining Social
Media Content for Crime Prediction,” in Web Intelligence
(WI), 2016 IEEE/WIC/ACM International Conference on,
2016, pp. 526–531.
[5] “Crimes - 2001 to present - Data.gov.” [Online].
Available: https://catalog.data.gov/dataset/crimes-2001-to-
present-398a4. [Accessed: 24-Jun-2017].
[6] M. Wang and M. S. Gerber, “Using Twitter for Next-
Place Prediction, with an Application to Crime Prediction,”
2015, pp. 941–948.
[7] X. Chen, Y. Cho, and S. Y. Jang, “Crime prediction
using Twitter sentiment and weather,” in Systems and
Information Engineering Design Symposium (SIEDS),
2015, 2015, pp. 63–68.
[8] W. Gorr, A. Olligschlaeger, and Y. Thompson, “Short-
term forecasting of crime,” Int. J. Forecast., vol. 19, no. 4,
pp. 579–594, 2003.
[9] T. Aljrees, D. Shi, D. Windridge, and W. Wong,
“Criminal pattern identification based on modified K-means
clustering,” in Machine Learning and Cybernetics
(ICMLC), 2016 International Conference on, 2016, vol. 2,
pp. 799–806.
[10] X. Zheng, Y. Cao, and Z. Ma, “A mathematical
modeling approach for geographical profiling and crime
prediction,” in Software Engineering and Service Science
(ICSESS), 2011 IEEE 2nd International Conference on,
2011, pp. 500–503.
[11] S. Sathyadevan, S. Gangadharan, and others, “Crime
analysis and prediction using data mining,” in Networks &
Soft Computing (ICNSC), 2014 First International
Conference on, 2014, pp. 406–412.
[12] A. Babakura, M. N. Sulaiman, and M. A. Yusuf,
“Improved method of classification algorithms for crime
prediction,” in Biometrics and Security Technologies
(ISBAST), 2014 International Symposium on, 2014, pp.
250–255.