We Have Numbers Of Free Samples

Abstract

Data analytics application in social and economic phenomena enables decision makers to better understand data, detect trends, explore relationships and drive insights that will be useful to take an efficient decision in future from historical data. This paper involves the exploration of crimes data by applying statistical techniques and machine learning algorithms, to discuss and analyses the claims of Ross Kemp and the Armed Police of 6th September 2018, regarding violent crime in the UK. These claims suppose that Violent Crime is increasing, there are more firearms incidents per head in Birmingham than anywhere else in the UK and that Crimes involving firearms are closely associated with drug.

1. Introduction

Statistics apply mathematics disciplines to analyze data. Most domains use data analysis to planning, analysis and interpret experimental work or historical data, especially for critical topics like crime and accident. Various existing studies have explored this type of data and look at the association between variables that capture attention to aid interpretation and presentation of results. Use the appropriate statistical method is essential to assess and make inferences in these critical topics; for example: the evaluation of the variation of Traffic Volume and Number of Accidents per Hour or the evaluation of the impact of round specificities in accident severity (GRANATO, 2014) (WANG, 2011).

There is various statistic techniques and methods proposed in the data analysis field. Descriptive statistics and inferential statistics are among the main statistical branches of statistics which are employed in scientific analysis of data and both are equally important for researchers in statistics.

Descriptive statistics summarizes data from a population sample using indicators like mean, median, minimum, maximum values, standard deviation and/or coefficient of variation. This type of statistics deals with the capture and presentation of data and often constitutes the first part of a statistical analysis. The statistician needs to be aware of designing experiments in this step, choosing the right focus group and avoid biases.

However, inferential statistics allow us to draw the right conclusions from the statistical analysis that has been performed using descriptive statistics. Among the main used techniques, we found Hypotheses testing and ANOVA (MILLER JR, 1997).

Two characteristics of data sets must be considered prior to the application of any inferential tests :1-Do the data conform to the principles of normality. i.e. normal distribution of data (ND)? 2- Do the data satisfy an assumption of homoscedasticity, i.e. uniformity of variance?

Most predictions of the future and generalizations of results to overall population using a smaller sample come under the umbrella of inferential statistics. By designing the right experiment and using the right techniques, the researcher is able to draw relevant conclusions to his study.

Hypothesis Testing is a statistical hypothesis is an assumption made by the researcher about the population of data collected for any experiment. This assumption can be true or false. Hypothesis testing is, the formal way of validating the hypothesis made by the researcher and check whether the hypothesis is accepted or rejected.

In order to validate a hypo (WANG, 2011) thesis, it will consider the entire population which is not possible in practice. So, to validate a hypothesis, we use random samples from a population and on the basis of the result from testing over the sample data, we can either select or reject the hypothesis (WILLIAM, 2000).

Statistical Hypothesis can be categorized into 2 types as below:

• Null Hypothesis –Hypothesis tests are used to test the validity of a claim that is made about a population. This claim that’s on trial, in essence, is called the null hypothesis. The null hypothesis testing is denoted by H0.
• Alternative Hypothesis –The alternative hypothesis is the one you would believe if the null hypothesis is concluded to be untrue. The evidence in the trial is your data and the statistics that go along with it. The alternative hypothesis testing is denoted by H₁or H_a.

Hypothesis testing uses a p-value to evaluate the strength of the evidence. The p-value is a number between 0 and 1 and interpreted in the following way (AL-GHAMDI, 2002):

• A small p-value (typically ≤0.05) indicates strong evidence against the null hypothesis, so you reject it.
• A large p-value (> 0.05) indicates weak evidence against the null hypothesis, so you fail to reject it.
• A p-value very close to the cutoff (0.05) is considered to be marginal and could go either way.

This study focus on the three kinds of statistic methods frequently used in this field as noted above, to practice and apply in crime case studies.

The purpose of this document is to present and apply data exploration, statistic techniques and machine learning, then discuss the outcome of applying them to our datasets.

The reminder of this paper presents two case studies. In each case study, it presents the dataset structure, applies at least 2 statistical techniques on it and discuss the results of using these techniques and if one supported more than others.

2. Overview of the Dataset

The U.K. crime dataset is compiled from reports sent by the 43 geographic police forces of England and Wales, the British Transport Police, and the Police Service of Northern Ireland. In addition, the Ministry of Justice supplements the crime reports with court outcomes data. This data is collated and submitted monthly by the regional police forces.

Each month, the regional police forces assemble a Crime and Anti-Social Behaviour (ASB) report file and submit this to central government. The Ministry of Justice then adds court outcomes from its records and references this against the police data as possible. All data is made anonymous before publication.

In this work we investigate the research questions using real, publicly available data sets that will be made available in Amazon S3. This dataset include:

• Street level crime data Published by the UK Home Office, this dataset contains at least 12 million data rows giving a crime type, together with their location as a latitude and longitude.
• POSTCODE DATA: This data set provided by the Ordinance Survey gives a latitude and longitude to every postcode. This is useful in the product to provide a relation between the Land Registry Price Paid dataset postcode, and the original crime dataset latitude/longitude and also indicates population in each postcode.

Research Question:

Prediction techniques are used in this research to discuss the following:

• Violent Crime is increasing,
• There are more firearms incidents per head in Birmingham than anywhere else in the UK
• Crimes involving firearms are closely associated with drug

Data Preparation and Cleaning

“Crime.csv” file Contains data of crimes from 2014 with multiple variables and a big volume of observations. Below is a snapshot of the selected dataset.

Figure 1:  Snapshot of crime data used columns

The data analysis of the crime data is conducted using big data platform based on the Apache Spark environment and python programming language too. The process of UK crime data is processed using Apache Spark. To explore relevant crimes data, we apply two main Filters to the dataset.

• The first filter include only crimes for which firearms offences are recorded as ‘Weapons’. However,
• The second filter concerns crimes that include a weapons offence together ‘sent to prison’ as an outcome since firearms possession carries a mandatory prison sentence.

After loading the main libraries in Python, the python script checks the schema and the main statistics of each data file using multiple summary function.

Exploratory Data Analysis

UK Crimes analysis using time series

The trends analysis of crime data using the count of number of crime over months shows that the number of crime increase over time which confirm the first research question and confirm the first claim.

Figure 2: UK crime trends by month

Exploration of UK Population

The analysis of population volume illustrates that Birmingham, is the most peopled city in UK which is relevant t study the correlation with crime number.

Figure 3: Population of top 20 UK cities

Volume of crime in Birmingham and Correlation with the number of crime

If we analyses the number of crime in UK cities, we observe in figure 4 that Birmingham is the top of UK cities in terms of crime volume. This claim can be explained by the population volume of his city which is the highest.

Given this last remark, we are interested by analyzing the correlation between the number of crime and the population of each UK city.

This previous figure presents the correlation between population of a c UK city and the number of crime of this city.

Figure 4: Correlation between crime and population

Figure 5: Top crime by city

Figure 6: 3D visualization of population and crime

Machine Learning Application

Predictive analytics is the technique of examining the past or historical data with the end goal to forecast the future result. The mostly used in the predictions are classification algorithms which are based on historical data. Classification is a one of prediction technique, given that it is supervised in nature. Classification technique possesses the capacity to predict the name for classes, provided that sufficient numbers of training data are provided in input. It exists multiple available classification algorithms, like Support vector machines, k Nearest Neighbors, weighted voting and Artificial Neural Networks. All these algorithms can be applied to a dataset for determining set of models to identify the unknown class category (MURRAY, 2015)

In classification techniques, the dataset has to be splatted into two sets, the first is the training dataset and the second is a testing dataset (independent set). Initially, the machine learning algorithm turns on the training data, then after the prediction model is applied on the test data. The following are some of the classification algorithms that are used in crime predictions.

Decision Tree Algorithm

To identify the weight of each variable on the crime volume, we propose to apply Decision tree algorithm. This classification technique uses decision trees to find the prediction value of the response variable. The decision trees are principally some functions allowing to predict and determine the class of the target variable while considering the different input variables (WIJENAYAKE, 2018).

There are multiple advantages of using the decision trees algorithms to predict target variable. Among them, the input of a decision tree classifier is tested beside a sub sets of the original data, obtained using some splitting rules or decision functions. This way helps to avoid excessive computations. The second advantage of this classification technique is that we can utilize a feature selection technique to ensure that we select the features meaningful to consider for the “decision tree classifier”. And finally, the efficiency of the decision tree algorithm is enhanced each time we reduce the number of selected.

Generally, a top down approach is applied to build a decision tree, until the predefined stopping criterion is met.

The decision tree (DT) presents a “hierarchical set of instructions which defines in what way dividing a big collection of data into sequentially reduced dataset (BRIJAIN, 2014).

• “With every succeeding division, the members of the ensuing segments become additional kind of like each other with relevancy the target”.
• DT utilizes the response variable to fix in what way the dataset must be divided by the input variables.
• DT is among the best powerful supervised data mining techniques, since, it can be used for a varied range of cases and it produces models that are easily comprehensive.

After the application of DecisionTree algorithm on Crime data, we found that the accuracy and test error of the decision tree model is illustrated below:

• Accuracy of DecisionTreeClassifier is = 0.830189
• Test Error of DecisionTreeClassifier = 0.169811

Linear regression to predict crimes volumes for each LSOACode

We have used Spark platform functions to apply the linear regression algorithm and try to predict the crime volume for each LSOA code. The used code is illustrated below:

In result, we have obtained the following:

• RMSE: 0.784348
• r2: 0.102906

RMSE measures the differences between predicted values by the model and the actual values. However, RMSE alone is meaningless until we compare with the actual “cnt” number of crime statistics, such as mean, min and max. After such comparison, our RMSE looks pretty good.

Discussion

The aim of this study was to analyze and explore UK crime data while discussing the claims of Ross Kemp and the Armed Police. Our analysis confirmed that these claims are right. Methods used for data analysis are statistical techniques, and machine learning algorithms: Decision tree and linear regression. A strong correlation has been identified between number of crime and the population size The study of the demographic structure, the population or the economic conditions of the areas can help to reveal the root cause of this relation and then the factors that impact the crime volume.