Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Exploratory Data Analysis And Crime Prevention

Using Machine Learning: The case of Ghana
Wellington Amponsah Parvinder Kaur
Computer Science and Engineering Er., Computer Science and Engineering
Shaheed Udham Singh College of Engineering and Shaheed Udham Singh College of Engineering and
Technology, Punjab Technical University Technology, Punjab Technical University
Mohali, India Mohali, India

Abstract:- The objective of this work is to take advantage I. INTRODUCTION

of machine learning to perform exploratory analysis of
historical data and to forecast crime counts in a given Artificial intelligence has over the years become the
month and year for a 4 year period between 2018 and subject of the day when issues of the future are brought to the
2021 and allow the evenly distribution and allocation of table. It comprises machine learning and deep learning which
resources and logistics in the case of Ghana. The all subset of the former. The application of Artificial
prediction was done using the Chicago crime dataset. The intelligence can be observed in the security agencies, self-
prediction was done using the Facebook prophet. The driving cars, chatbots, in the financial sector, the health and
month February is the month with the least crime rate agriculture sector, among others. The increase in population
and this can be attributed to the fact that it has fewer days contributed by migration and high birth rate has become one
in the year. It was also discovered that crimes are of the defining themes in the twenty-first century in Sub-
committed between the hours of 5:00pm and 10:00 pm Saharan Africa. Increasingly as populations explode and life-
while most of the crimes are committed at 12:00 noon. spans increase due to increased accessibility and availability
With regard to District level crimes, it was observed that to better health care, technology and education, there is
District 11 is the district with the highest crime between pressure on the security agencies and the resources required
2012 and 2017. This is followed by 7, 4, 25 and 6. This is sustaining the security of the country. Of great significance is
an indication that more logistics and personnel will be how the security of the state and its citizens and other
required in those Districts to help prevent crimes from immigrants are managed optimally. This has also led to the
being committed. forecasting of crimes and the ability to predict crimes using
all forms of statistical methods and tools to identify hotspots
The model predicted a decrease in the number of in the fight against crime. The police administration is the
crimes that are likely to be committed with 2021 institution in Ghana that is responsible for the security of the
recording the least crimes. The model also predicted the citizens. [1] The Ghana Police Service (GPS) is the main law
least crimes to be in the period 1st to 30th January, 2021 enforcement agency of Ghana. The service is under the
as 10978 as compared to 1st to 30th January, 2021with the control of the Ghanaian Ministry of the Interior and employs
least crimes committed as 30 in the historical data. Most over 30,000 officers across its 651 stations. [2] The primary
crimes happen on street and on sidewalks therefore extra functions of the Service are those stipulated in Section 1 of
police personnel needed on street patrolling. A lot of the Police Service Act, 1970 (Act 350). It states among others
crimes are in residence and/or apartments therefore the that: “It shall be the duties of the Police Service to prevent
Police Service will require more personnel to respond to and detect crime, to apprehend offenders and to maintain
destress 911 calls from people. The overall trend is that public order and safety of persons and properties”. Each year
the crime rate keeps decreasing from the forecast in each the police administration submits a budget to the ministry of
year. The results indicate the importance of the finance and large sums of monies are appropriated to
application of Machine Learning for the prediction of maintain law and order. It is also important to acknowledge
crime data by the Ghana Police Service. In conclusion, that we live in a world of data that every activity we do
this work provides the institution with much information generates some amount of data. And these data can be
and intuition on the use and application of machine leveraged in the prediction of crime and their subsequent
learning to enhance the decision making process and the prevention. The efficiency of the security agents is their
fight against crimes. ability to quickly use the available data to investigate crime
and also to take crime prevention actions by the leverage of
Keywords:- Ghana Police Service, Machine Learning, machine learning algorithms which will be one of the
Artificial Nueral Network, Predictive Policing. efficient ways to boost the effectiveness of the police
department through optimization of their operations. The
purpose of this paper is to examine how machine learning
algorithms can be used explore crime data and to predict
crime seasons while identifying areas of high probability of
crimes being committed.

IJISRT21MAR509 www.ijisrt.com 613

Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
II. SURVEY OF LITERATURE crime data. It is able to visualize and present a graphical
analysis of crime patterns in the society and categories crimes
A. Machine Learning based on their distinct related types. It uses the Formal
[3] S.N.Sivanandam et al, The term "machine learning" Concept Analysis which is a data analysis technique for
is today primarily associated with attempts to get computers to considering the common and distinct attributes of crimes such
solve problems that use a particular small and somewhat that they have relations. The relationships between crimes,
arbitrarily chosen set of methodologies. It is a subset or branch their location and occurrence were visualized using the
of artificial intelligence performing actions without it being formal concept analysis techniques. In their paper, crime
explicitly programmed or instructed by learning from a types were considered and the relationship that a particular
historical or previous data. Machine learning focuses on the crime belonged to a particular geography and estimated
development of algorithms that are able to solve problems however, the particular crime that could occur within that
through the process of learning by getting better over the task particular geography was not established.
as they learn on the data over a period with experience given a
certain number of epochs. These algorithms are able to learn Obubu Maxwell, et al [11] examined murder crimes in
to perform the task they are intended to do using developed Nigeria using the Autoregressive integrated moving average
models for predictions over unseen datasets using important model and also forecasted crimes using historical data. They
data. [4]Machine Learning uses data mining techniques to used an exploration method to perform a prediction using a
extract the information from the huge size datasets. Machine historical time series data. In their paper, the autoregressive
learning can be classified into two broad classifications integrated moving average was used to forecast the murder
namely supervised learning and unsupervised learning. [5] crime and the increasing murder crimes. However, other
There are five types of machine learning algorithms that are models or algorithms were not used in the crime forecast to
used on the datasets. These algorithms are; Classification determine the accuracy of the model. Their model will not be
Analysis Algorithm, Regression Analysis Algorithm, able to be use to perform a future occurrence of murder
Segmentation Analysis Algorithm, Sequence Analysis crimes. It also did not consider other important crimes which
Algorithm and Association Analysis Algorithm. equally affects the livelihood of the people of the society

B. Artificial Nueral Network Luiz G.A Alves, et al [12] examined the crime dataset
Artificial Neural Network is also called neural network. of Brazil. They performed predictive analysis of crimes
S. N. Sivanandam et al [6] an artificial neural network (ANN) dataset and tried to obtain the relationship between crime and
may be defined as an information processing model that is urban indicators. They proposed that due to the non-Gaussian
inspired by the way biological nervous systems, such as the distribution and multicollinearity in the urban indicators, it
brain, process information. This model tries to replicate only makes it very easy to find conclusion about the influence if
the most basic functions of the brain. The key element of some urban indicators on crime. Random forest algorithm
ANN is the novel structure of its information processing was used to predict crimes and the influence of urban
system. An ANN is composed of a large number of highly indicators on crimes such as homicides. It also took into
interconnected processing elements (neurons) working in consideration the ranks of the various indicators and predicted
unison to solve specific problems. The fundamental data that unemployment and illiteracy are the most important
structure in neural networks is the layer which is the data- variables for homicide crimes in cities in Brazilian.
processing module. This data processing module accepts one
or more tensors as inputs. Some layers are stateless, but more However they could not predict or forecast which
frequently layers have a state: the layer’s weights, one or crimes are most likely to be committed in a given period. It
several tensors learned with stochastic gradient descent, only took the accuracy of the model which it says can predict
which together contain the network’s knowledge. A neural accurately at 97% of a crime based on some attributed and
network has its main responsibility of learning during the indicators that have effect on crime to guide crime control. It
process of training and it is able to adapt to stimulus. This is is also important to note that their model was able to perform
through the [7] parameterization of the weight where the prediction with accuracy between 38% and 39%.
weight is consistently adjusted until it is able to produce the
desired target or response. The main components of neural Rizwan Iqbal, et al [13] used the classification
networks are the layers, networks, objective functions, and algorithm to predict crime for different states in the United
optimizers. [8] Proposed a dynamic programming based States of America using real data. An open source tool called
recursive algorithm to find the similarity between the training WEKA developed in JAVA was used. The ‘Crime and
and test images. For ranking the images on the basis of Community’ dataset was used in their work. It performed the
highest similarity with the given object template a greedy results of two machine learning algorithms namely Naïve
multistate gradient descent method was used. The method Bayesian algorithm and Decision Tree algorithm to predict a
proved robust to rotation and deformations but the method particular category of crime. A 10 fold cross validation was
was not able to deal with extreme viewpoint changes. applied to the dataset separately for Naïve Bayesian algorithm
and Decision Tress algorithm. It then concluded that Naïve
C. Related Work Bayesian algorithm is the best for performing an accurate
Propose a new way of analyzing crime patterns using prediction of crime category at 83.9519 %. The researchers
the combination of Formal Concept Analysis and could not evaluate the performance of their prediction. It did
Geographical Information Science to discover patterns in not consider other features to study their effects on crimes

IJISRT21MAR509 www.ijisrt.com 614

Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
and what effect the features will have on the prediction of domain experts has helped the Ghana Police Service
crimes using other machine learning algorithms. however; it will not be able to determine some deviations that
are more likely to occur than machine learning can. As Ghana
Suhong Kim, et al [14] created a predictive model that is getting prepared to go to the polls on 7 th December, 2020,
was able to accurately predict crime. Two main algorithms crimes relating to elections which are commonly referred to
were used in their work. These are K-Nearest Neighbor and as electoral violence or political violence are highly
Decision tree algorithms. It also used the geographic anticipated to go on the rise. Electoral violence or political
information system to identify the location of crimes by the violence is a pervasive trait of the electoral period or election
leverage of the co-ordinates which can easily be located on year in the county. These actions demand that the Ghana
the map. However it is important to know that KNN Police Service become very proactive by predicting such
algorithms cannot be applied to a large datasets since it will crimes and where it would occur often by the use of a model
cost to be able to calculate the distance between the data that has a proactive power to be able to perform accurate or
points and also with sensitivity to noise and missing data in high predictions. With Ghana gradually becoming more
the dataset. The accuracy of the predictions using K-Nearest industrialized partly due to the recent oil found and the
Neighbor is absolutely dependent on the number of neighbors industrialized approach taken by the government, an
while the speed of the algorithm is declines more quickly. anticipated pressure on the security services to provide
security to citizens and investors is very necessary. It is the
III. POBLEM FOMULATION responsibility of the Ghana Police Service to be proactive as
well as their ability to allocate resources where necessary.
Companies and organizations are collecting data on According to [15], Predictive analysis is a complex process
their customers, employees, etc. which are stored in a digital that uses large volumes of data to forecast and formulate
form. The metrological department of Ghana collects data potential outcomes of an event. In a criminal jurisprudence,
about the weather. These weather data collected are cleaned, the work of estimating and combating crimes rests mainly
stored, and processed to make decisions on the weather and with police, probation practitioners, and other professionals,
crimes. They are able to forecast the weather condition for the who must gain expertise in the subject area over many years
coming days using predictive algorithms and other statistical and it is time-consuming and subject to bias and error. [16]
tools and technologies. These enable a timely and efficient The causes of crimes differ from country to country due to
decision and also speed up the decision making process. The different culture, economic and social characteristics and
Ghana Police Service as well gather data on criminal have therefore divided causes of the crimes primarily relating
activities across the country about the committed crimes, to economic situation, social environment and family
location of the crime, the type of crime, the names of victims structure. [17] In Ghana, crime is not limited to poor areas
and criminals if any, whether arrests were made or not, etc. and this is where the relationship becomes unclear. [18] We
Huge sums of monies are allocated to the Ghana Police urgently need a robust crime database that will complement
Service by the central government yearly in the annual budget our increasing use of intelligence to fight crime. Predictive
of the country to enable the police administration perform policing and Prompt identification of suspected criminals
their functions effectively. Monies are spent by the police through the application of artificial neural networks will
administration to forecast crime and determined hotspots. increase the efficiency of police operations in order to better
Crime has become one of the major threats that are facing all protect citizens against both internal and external threats.
countries of which Ghana is no exception. The perpetrators of Given that many individuals and organizations and
crimes have made it a necessary evil for various scholars to institutions are being established, the number of crime rate
research into crime and the behaviour of criminals in the increases, researching into the use of machine learning and
society in which we find ourselves. It has also warranted the deep neural networks in combating crime is paramount.
study of the application of predictive algorithms to enable the Crimes are committed on a daily basis and the security
security service predict the occurrence of crimes. service has over the years been reactive and has relied on
people for information on where, when and how crimes are
It is no doubt to know that crime prevention and being perpetuated. The security agency has over the years in
prediction manually has become a very complex activity by the fight against crime use hotspots based on knowledge
the Ghana Police Service and other security agencies leading obtained from previous data. The hotspots are used by the
to the large volume of crime data that is generated each year. Ghana Police Service to deploy their men-in-uniform where it
Time and resources are daily spent at a high rate to make the has been established based on the historical data that the
police service combat the high rates of crimes and to likelihood of crimes being perpetrated at these areas is high.
accurately predict the amount of crimes to occurred in a given According to [14], a hotspot is an area that has the highest
period and crime scenes through the studying of historical rate of crime during the forecast which is regarded as the
data. The trends and the correlation in the historical data are most dangerous area. This method is employed by the Ghana
used to estimate the amount of crimes to be committed in a Police Service which does not help in the fight against crime,
given period of time, what type of crimes are likely to be making the service reactive rather than proactive. It is
committed within a particular period, the prevailing crimes, important to know that as the crime rate in the country
the time and the day for a particular crime to be committed increases, the Ghana Police Service will need to adopt
through the application of statistical knowledge. This method artificial neural network and machine learning techniques by
of estimating crimes and crime scene using historical data and developing a predictive model to help in the fight against
the statistical tools and analytics tools and knowledge by crime.

IJISRT21MAR509 www.ijisrt.com 615

Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Predictive policing applying artificial neural networks Number equations consecutively. Equation numbers,
and machine learning are currently not utilized by the GPS. within parentheses, are to position flush right, as in (1), using a
The method of forecasting based on historical data is used to right tab stop. To make your equations more compact, you
forecast the likelihood of a crime being committed at a may use the solidus ( / ), the exp function, or appropriate
particular place. exponents. Italicize Roman symbols for quantities and
variables, but not Greek symbols. Use a long dash rather than
How can the security agency leverage machine learning to a hyphen for a minus sign. Punctuate equations with commas
predict crime? How can the security agency leverage machine or periods when they are part of a sentence, as in
learning to fully allocate resources to fight crime?
How can the security agency use machine learning to
discover crime pattern in the society?
How can the security agency apply machine learning to
optimize their operations in crime combat? Note that the equation is centered using a center tab stop.
How can the Ghana Police Service also utilize the biometric Be sure that the symbols in your equation have been defined
data from the NIA to extract the full data about a criminal before or immediately following the equation. Use “(1),” not
through the use of an image recognition model? “Eq. (1)” or “equation (1),” except at the beginning of a
sentence: “Equation (1) is ...”
Hence the use of machine learning algorithms and
artificial neural network techniques will be applied in this
work to the Chicago crime datasets comprising of criminal
activities to determine various what crimes are likely to be D. Some Common Mistakes
committed, the various months in which crimes are  The word “data” is plural, not singular.
committed, the years that crime rate are likely to either  The subscript for the permeability of vacuum 0, and other
increase and reduce, and the various clusters of crimes and common scientific constants, is zero with subscript
factors that could cause the decreased or increase in crime rate formatting, not a lowercase letter “o.”
at a particular time period.  In American English, commas, semi-/colons, periods,
question and exclamation marks are located within
A. Abbreviations and Acronyms quotation marks only when a complete thought or name is
Define abbreviations and acronyms the first time they cited, such as a title or full quotation. When quotation
are used in the text, even after they have been defined in the marks are used, instead of a bold or italic typeface, to
abstract. Abbreviations such as IEEE, SI, MKS, CGS, sc, dc, highlight a word or phrase, punctuation should appear
and rms do not have to be defined. Do not use abbreviations in outside of the quotation marks. A parenthetical phrase or
the title or heads unless they are unavoidable. statement at the end of a sentence is punctuated outside of
the closing parenthesis (like this). (A parenthetical
B. Units sentence is punctuated within the parentheses.)
 Use either SI (MKS) or CGS as primary units. (SI units are  A graph within a graph is an “inset,” not an “insert.” The
encouraged.) English units may be used as secondary units word alternatively is preferred to the word “alternately”
(in parentheses). An exception would be the use of English (unless you really mean something that alternates).
units as identifiers in trade, such as “3.5-inch disk drive.”  Do not use the word “essentially” to mean
 Avoid combining SI and CGS units, such as current in “approximately” or “effectively.”
amperes and magnetic field in oersteds. This often leads to  In your paper title, if the words “that uses” can accurately
confusion because equations do not balance dimensionally. replace the word using, capitalize the “u”; if not, keep
If you must use mixed units, clearly state the units for each using lower-cased.
quantity that you use in an equation.  Be aware of the different meanings of the homophones
 Do not mix complete spellings and abbreviations of units: “affect” and “effect,” “complement” and “compliment,”
“Wb/m2” or “webers per square meter,” not “webers/m2.” “discreet” and “discrete,” “principal” and “principle.”
Spell units when they appear in text: “...a few henries,” not  Do not confuse “imply” and “infer.”
“...a few H.”  The prefix “non” is not a word; it should be joined to the
 Use a zero before decimal points: “0.25,” not “.25.” Use word it modifies, usually without a hyphen.
“cm3,” not “cc.” (bullet list)  There is no period after the “et” in the Latin abbreviation
“et al.”
C. Equations  The abbreviation “i.e.” means “that is,” and the
The equations are an exception to the prescribed
abbreviation “e.g.” means “for example.”An excellent style
specifications of this template. You will need to determine
manual for science writers is [7].
whether or not your equation should be typed using either the
Times New Roman or the Symbol font (please no other font).
To create multileveled equations, it may be necessary to treat
the equation as a graphic and insert it into the text after your
paper is styled.

IJISRT21MAR509 www.ijisrt.com 616

Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IV. SOURCES OF DATA VII. TRAINING OF THE ALGORITHMS

A secondary data is the type of data that has already After the data has been cleaned, exploited and
been collected by means of a primary source and it’s been visualized, it is time to select the machine learning model.
made available for researchers to use for their own research. Selecting or choosing a machine learning model enables us to
This data may be collected for general use with absolutely no perform the work easily and also to obtain the right
specific research purpose. Sources of data which are predictions we expect. Some of the machine learning models
classified as secondary data sources include books, personal includes Classification, Regression, Dimensionality
sources, journal website, government record, non- Reduction and Clustering. The choice of the model to use
governmental organizations, and newspapers among others. shall depend on the amount of data available for the purpose
The prevalence of the internet and the electronic media has of the training and testing and prediction. It is also very
made access to secondary sources of data easy without any paramount to note that the choice of model selection also
difficulties. It saves time without spending much time in depends highly on the problem to be solved. In machine
collecting data, providing cost efficiencies. It serves as the learning a model is created in order to predict the outcome of
baseline for the primary research helping in research design the event such as predicting the crime rate, the price of a
and is associated with a quantitative database serving as a house among others. After the model is created, the
legitimate avenue for quantitative research. performance of the model is calculated using the method
called Train and Test.
In this research work, a secondary source of data will be
the basis. Secondary sources will be used to collect all data VIII. EVALUATION OF THE MODEL
relevant for the study. This secondary data for the research
work will be the Chicago Crime datasets. The Chicago Crime The important part of performing a predictive modeling
datasets will be the primary focus of this work however if is evaluating the model. A machine learning model is always
possible data on crime activities on rural and farming evaluated to determine how good the model is at predicting
households and their accessibility to state security in the the target of the new and future data. Model evaluation aims
event of a commitment of a crime may be considered and at estimating the general accuracy of a model on future data.
shall be stated wherever applicable The methods for evaluating a model’s performance are
divided into two main categories namely, holdout and Cross-
V. DATA COLLECTION AND PREPARATIONS validation methods. Cross-validation is also called out-of-
sample testing used to determine how well the result of model
Data collection is the process by which data is gathered generalizes well to an unseen dataset. It involve partitioning
or collected through relevant sources to find solutions to a the original observation into training set which is used for
research problem, to test a hypothesis of the research and training purpose and an independent set which is used to
evaluate the outcomes. Data collection is an important part of evaluate the performance of the model. The most common
quantitative research which captures quality evidence cross validation technique is called k-fold cross validation.
allowing analysis to lead to the formulation of a credible The holdout validation is to test the model on a different data
solution to a problem. than the data it was trained on to provide an unbiased estimate
of the learning performance. It involves the random division
A dataset is a container for our data storage which is into three subset namely Training set, validation set and the
mostly presented in a two dimensional array form. It is made test set. Both methods use a test set to evaluate model
up of series consisting of some rows and columns. The performance. It’s not recommended to use the data used to
dataset for this work involve 22 attributes. The data is build the model to evaluate it. The reason being that the model
preprocessed using data mining techniques to transform the will simply remember the whole training set, and will
raw data into an efficient, useful and meaningful format. therefore always predict the correct label for any point in the
training set which will result in overfitting of the model.
VI. DATA VISUALIZATION
IX. EXPLORATORY DATA ANALYSIS
Data visualization is the means of conveying
information to users in a graphical view. This can be in the The entire dataset is visualized and analyzed. And
form of graph, maps, etc. I can be said to be a graphical missing values were all dropped from the dataset before
representation of information for easy understanding and prediction. Seaborn and matplotlib were used for the purpose
interpretations. The important information in the dataset is of visualization.
exploited to obtain good information about the Chicago crime
dataset. This is done through the process of Exploratory Data
Analysis (EDA). This enables us to see some useful trends in
the dataset.

IJISRT21MAR509 www.ijisrt.com 617

Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
which crimes are committed most since it has the highest
number of crimes committed and Saturday has the least.

Figure 1 Figure 2

From the dataset between 2012 and 2017, the number of X. LOCATION OF CRIMES
domestic crimes committed were 12000000 and 200000
represents the total number of other crimes. The history data was visualized to present the
description of crimes that are committed and the frequency.
A total number of 110000 offenders could not be From the diagram below, it is observed that the top four
arrested while about 40000 arrests were made by the police crimes are committed on STREET, RESIDENCE,
department. This is represented graphically below. APARTMENT and SIDEWALK. Other crimes are
committed on OTHER, PARKING LOT/GARAGE, ALLEY,
RESIDENTIAL YARD, SMALL RETAIL STORE and
SCHOOL, PUBLIC and BUILDING. Street crimes are the
most committed crimes which records 330471 incidents. The
least crime is committed in School, Public and Building.

Figure 2

The data was studied to observe the top ten (10) crimes
that are committed with theft being the most crimes to
robbery in the topmost crimes. A little over 300000 thefts
were committed in the between 2012 and 2017 with robbery
of a little over 50000 crimes in the years as the least crime
among the top 10 crimes between 2012 and 2017.
Figure 1
A top 10 crimes that were committed is shown below in
the graph From the exploratory analysis being done on the
historical data, the location of crimes is exploited to
determine the District with the highest crimes. From the
diagram below, the lowest amount of crime is committed in
the District 13. District 11 recorded the highest crime
between the year 2012 and 2017.

Figure 1

From the analysis from the below figure 4, the number Figure 2
of crimes committed appears evenly distributed across days
of the week. We cannot therefore say that crimes are more From the below plot in figure 7, it can be inferred that
likely to be committed on a particular day of the week more crimes were committed between May and August of
however, it is imperative to assume Thursday to be the day which July has the highest number of crimes. The month of

IJISRT21MAR509 www.ijisrt.com 618

Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
February has the lowest number of crimes which can be as a evaluation. Making predictions or estimations on historical
result of the fact that it has lesser days in the year. Figure 5.8 data of the likelihoods about future events is what is called
and figure 9 also indicates crimes that are committed per day predictive analysis. It gives us more knowledge into the
graphically from the year 2012 to 2017. future based on probability and machine learning and
artificial intelligence. Machine learning focuses on predictive
analysis through the combination of historical data to make
inferences.

An example of using Facebook’s Prophet Tool to

forecast the future crime for the crime dataset per month is
shown below in table 2.

Table 1
S/N Dates Actual Predicted
Crimes(y) Crimes(yhat)
1 2014-02- 17684 17892
Figure 3 28
2 2014-03- 19155 22005
31
3 2014-04- 20949 22519
30
4 2014-05- 22619 24413
31
5 2014-05- 20203 24867
31

According to the historical data, the highest crime was

committed in the year 2012 with a total number of crimes as
Figure 4 334397 and the year 2017 recorded the lowest in terms of the
crimes committed. This is shown in the figure below.

Figure 5
Figure 1
XI. THE PROPET
The model was able to predict the total number of
The Prophet is a tool developed by Facebook for the crimes that could be committed in months for 1460 days
purpose of time series forecasting. It is one of the tools that representing 4 years based on the historical data that was
are capable of predicting at a reasonable quality. It is an open- obtained. This is from the year 2018 to 2021. The model
source tool based on an additive model where nonlinear predicted a decrease in the number of crimes that are likely to
trends are fit with yearly, weekly and daily seasonality. It also be committed with 2021 expected to record the least crimes.
has holiday effects. It is used in many applications for the It is important to note that the number of crimes keeps
production of reliable predictions. declining each year since 2012.
In this work, the Prophet is used for the purpose of The model also predicted the least crimes to be in the
modeling the dynamics of crimes. This is done to enable the period 1st to 30th January, 2021 as 10978 as compared to 1st
generation of daily, weekly, monthly and quarterly crime to 30th January, 2017with the least crimes committed as 30 in
predictions. the historical data. The table below is the forecast
components which plot the trend, yearly and weekly
In application of the Prophet, it is very important that at seasonality of the Chicago crime datasets between 2012 and
the least a 24 month of the historical data is made available 2021. We therefore visualize the prediction of the crime from
for the efficient and reliable estimation of the various trends. the year 2018 to the year 2021.
A quality prediction is very important after the model

IJISRT21MAR509 www.ijisrt.com 619

Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

(2)

-Root Mean Squared Error is another very good performance

metric which is the variation of the Mean Squared Error. It
determines the mean deviation that is in the predicted values
from the actual values. It is based on the assumption that error
is unbiased following a normal distribution. It is defined
mathematically as:

(3)
Figure 2
The table below is the metrics of our model for a given
From the diagram below, the blue line in the graph period or horizon using the facbook prophet.
represents the predicted values which are the prediction and
the black dots represent the data in the historical dataset. Table 1
Horizon MSE RMSE MAE
48 1.606 12.67 8.35
51 1.600 12.65 8.17
53 1.74 13.22 9.45
56 1.64 12.81 8.52
59 1.90 13.79 10.49

XIII. EVALUATING OUR MODEL

This includes the selection of the measure as well as the

actual evaluation; seemingly a smaller step than others, but
important to our end result. It is as important as the quality of
our predictions since it helps to know how well the models
Figure 3 perform in real world situations.
XII. MODEL PERFORMANCE METRICS The model evaluation is done after the training of the
model using a cross validation. In machine learning, our
The performance metrics utility can is used to compute dataset is trained and the model is generated which is
useful statistics of the prediction performance as a function of compiled. After the training, the model is then evaluated to
the distance of how far into the future the prediction was. see its performance on the training data and the test data. The
validation is the process by which a part of the training data is
Three main performance metrics are computed in this aside for the validation of the model accuracy. Cross
work namely Mean Squared Error (MSE), Root Mean validation procedure is used to validate the model. This cross
Squared Error (RMSE) and Mean Absolute Error (MAE). validation procedure is done automatically for a range of
historical cutoffs using the cross validation function provided
-Mean Squared Error is the average of the square of the in the Facebook prophet module. The horizons, the initial
distance between the actual values and the predicted values. period and the spacing between the periods are specified.
This is defined mathematically as;
A data frame as output is provided by the cross
validation process. This presents us with the true values
represented, ‘y’ and the predicted values, ‘yhat’ each for a
(1) predicted date and the cutoff date. The prediction is made for
each of the observed data points between cutoff and the cutoff
-Mean Absolute Error is the average of the difference between horizon. The ‘y’ and ‘yhat’ is used to calculate the error or
the Original Values and the Predicted Values. It gives us the the difference between the actual values and the predicted
measure of how far the predictions were from the actual values to enable us obtain the residual values of our model.
output. However, they don’t give us any idea of the direction The cross validation was performed to compute the
of the error i.e. whether we are under predicting the data or performance of the prediction for a period of 365 days. The
over predicting the data. Mathematically, it is represented as cross validation began with 730 days of training data in the
initial cutoff and predictions are performed after every 30

IJISRT21MAR509 www.ijisrt.com 620

Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
days. The figure below depicts the cross validation process predictions and also apply it in detecting crimes, its
for a period of 30 days. occurrence, where there is high probability of crime to be
committed. This will also enable the service to be able to
After the cross validation process, an output is obtained. allocate more resources to place where much resource will
The content of the cross validation is the dates, the actual likely be needed. In conclusion, this work provides the
values (y), the predicted or forecast values (yhat), the institution with much information and intuition on the use and
uncertainty intervals namely the yhat_lower and yhat_upper application of machine learning to enhance the decision
and the cutoff point. making process and the fight against crimes.

XIV. RESULTS ACKNOWLEDGMENT

The forecast or prediction is evaluated on real-world I express my sincere gratitude to the Almighty God,
data. More resources should be allocated to the police Shaheed Udham Singh College of Engineering and
department to fight crime. It can be concluded that most Technology, I.K.G Punjab Technical University, Jalandhar
crimes occur on the street, residence followed by school, for giving me the opportunity to work on the thesis during my
public and building and therefore, most police personnel must final year of M.Tech. I owe my sincerest gratitude towards
be deployed to these areas. Much resources and logistics must Professor Sushil Kamboj, ER Parvinder Kaur, Er Deepinda
be allocated to fight crimes in these locations. According to Kaur for their valuable advice and healthy criticism
the exploratory data analysis on the historical data, it can be throughout my thesis which helped me immensely to
concluded that most offenders are not arrested of crimes that complete my work successfully. I would like to thank Head
they have committed. Much as 110000 offenders were not of Department CSE and members of the Departmental
arrested while only 40000 offenders of various crimes were research team for their valuable suggestions and healthy
arrested. This could be due to the wrongly allocation of criticism during my presentation of the work. I would also
logistics to places with less crimes. On the average, more like to thank Professor Sukhpreet Kaur, ER Rasneet Kaur, Dr.
crimes are committed between the month of May and August Manavjot, and all my lecturers at the CSE department and
each year. The month February is the month with the least SUS College for their help. I would also like to thank my
crime rate and this can be attributed to the fact that it has parents, Sylvia Somuah, Nancy Edu, Mavis Darko, Andrews
fewer days in the year. Oberko, Asare Lydia, Taiwo Emmanuel, Mary Margaret
Dakurah, African students of SUS, friends, etc. who helped
It was also discovered that crimes are committed me one way or the other in the process of my studies.
between the hours of 5:00pm and 10:00 pm. However it is
very imperative to understand from the data that most of the REFERENCES
crimes are committed at 12:00 noon during the day.
[1]. https://en.wikipedia.org/wiki/Ghana_Police_Service,
With regard to District level crimes, it was observed Access online on 5 March 2020
that District 11 is the district with the highest crime between [2]. https://police.gov.gh/en/index.php/functions/# Access
2012 and 2017. This is followed by 7, 4, 25 and 6. This is an online on 5 March 2020
indication that more logistics and personnel will be required [3]. S.N.Sivanandam and S.N. Deepa, “Principles of Soft
in those Districts to help prevent crimes from being Computing”
committed. [4]. Rekha Nagar1, Yudhvir Singh, U.I.E.T (M.D.U), “A
literature survey on Machine Learning Algorithms,”
The overall trend is that the crime rate keeps decreasing [5]. Lawrence McClendon and Natarajan Meghanathan,
from the forecast in each year. We now know that most “Using machine learning algorithms to analyze crime
crimes happen on street and on sidewalks therefore we need data”.
extra police personnel on street patrolling. A lot of crimes are [6]. S.N.Sivanandam and S.N. Deepa, “Principles of Soft
in residence and/or apartments therefore the Police Service Computing”
will require more personnel to respond to destress 911 calls [7]. S.Agatonovic-Kustrin, R Beresford “Basic concepts of
from people. artificial neural network (ANN) modeling and its
application in pharmaceutical research”
XV. CONCLUSION [8]. K. Schindler, D. Suter, “Object Detection by Global
Contour Shape”, Science Direct Journal on Pattern
In this paper, Facebook Prophet was used by applying Recognition (2008), Vol. 41, Issue 12, pp. 3736–3748
machine learning. The model used Facebook prophet to [9]. Alice Zheng et al Amanda Casar, “Feature Engineering
predict or forecast crime. It is recommended for the Ghana for Machine Learning Principles and Techniques for
Police Service and other security agencies to apply machine Data Scientists”.
Learning and artificial Intelligence in the combat of crimes in [10]. Kester Quist-Aphetsi ‘Visualization and analysis of
the country. The Police Service can use this method to geographical crime patterns using formal concept
forecast crimes in the country. The amount of crime to be analysis” IJRSG ISSN No: 2319-3484 Volume 2, Issue
committed in the year and month. It is hereby concluded that, 1, Jan. 2013.
it is very necessary to ensure that Ghana Police service and
other security agencies apply Machine Learning in crime

IJISRT21MAR509 www.ijisrt.com 621

Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[11]. Obubu Maxwell, Ikediuwa U.Chinedu, Anabike
C.Ifeanyi and Nwokike Chukwudike, “On Modeling
Murder Crimes in Nigeria” ISSN: 2687-8097 DOI:
10.33552/ SJRR.2019.02.000528 Volume 2 - Issue 1,
2019.
[12]. Luiz G. A. Alves, Haroldo V. Ribeiro, Francisco A.
Rodrigues “Crime prediction through urban metrics and
statistical learning”, Physica A 505 Volume 505, 1
September 2018, Pages 435-443
[13]. Rizwan Iqbal, Masrah A.A Murad, Aida Mustapha,
P.H.S Panahy, and Nasim Khanahmadliravi, “An
Experimental Study of Classification Algorithms for
Crime Prediction”, IJST
DOI: 10.17485/ijst/2013/v6i3.6, 2013, Volume:
6, Issue: 3, Pages: 1-7
[14]. Suhong Kim , Param Joshi, Parminder Singh Kalsi, and
Pooya Taheri, “Crime Analysis Through Machine
Learning”, Fraser International College, Simon Fraser
University British Columbia, Canada 2018
[15]. Crime and the “poverty penalty” in urban Ghana, The
Safe and Inclusive Cities initiative, University of
Ghana, international development research Centre.
[16]. Ali Muhammad, “Youth Crime: Causes and Remedies”,
MPRA Paper No. 17223, posted 11 Sep 2009 06:31
UTC
[17]. https://www.ghanaweb.com/GhanaHomePage/NewsArc
hive/Ghana-Police-securityservices-builds-pre-emptive-
Central-Crime-Database-IGP-702640
[Accessed:01/02/2020]
[18]. Robert Bogucki, Jan Kanty Milczek, Patryk Miziula “A
simple crime hotspot forecasting algorithm”, fedcsis, pp.
23–26 DOI: 10.15439/2020F5 ISSN 2300-5963 ACSIS,
Vol.

IJISRT21MAR509 www.ijisrt.com 622