Youssef Taher,Fouad Moussaoui and Brahim Lamrabti,Amal Azaroual, Kawtar Bouzoubaa, Hanaa Moussaoui
Center of guidance and planning of education -Rabat -Morocco
In the 21st century, and across the world, fighting crime is major concern, particularly the assessment of recidivism risk factors (the likelihood of a person convicted of a crime to offend again). Today, some criminal justice systems starting to use big data to predict how likely someone who has just been released from prison is to commit another crime within a few years. The main objective is to predict where and when crime will happen before it actually occurs. These new predictions of recidivism are often solved by dozens software based on data- driven approaches. These software solutions are limited in terms of cases handled and accuracy. In this context, and by combining theory-driven and data-driven approaches, the current work proposes an example of big data conceptual framework oriented to recidivism prediction. To predict, analyze and compare different possible cases of recidivism, we aim for integrating all three types of recidivism (general, violent, and sexual recidivism) in one design, and optimizing the critical features to predict accurately each type. The proposed framework is based in three important modules to integrate the prediction of the three types of recidivism, and three important theoretical digital libraries to support the decisions making by the data-driven approach.
Recidivism, conceptual framework, big data, recidivism prediction
Gaseb, Alotibi1 Nathan Clarke2, Fudong Li2 and Steven Furnell2,3, 1Ministry of Interior, Public Security, IT Department Plymouth University, Plymouth, United Kingdom, 2Security Research Institute, Edith Cowan University, Western Australia and 3Center for Research in Information and Cyber Security, Nelson Mandela Metropolitan University, Port Elizabeth, South Africa
Insider threat has become a serious issue to the many organizations. Various companies are increasingly deploying many information technologies to prevent unauthorized access to getting inside their system. Biometrics approaches have some techniques that contribute towards controlling the point of entry. However, these methods mainly are not able to continuously validate the users reliability. In contrast behavioral profiling is one of the biometrics technologies but it focusing on the activities of the users during using the system and comparing that with a previous history. This paper presents a comprehensive analysis, literature review and limitations on behavioral profiling approach and to what extent that can be used for mitigating insider misuse.
insider threat, behaviouial profiling, insider misuse
Befekadu G. Gebraselase, Bjarne E. Helvik, Yuming Jiang, NTNU – Norwegian University of Science and Technology
The underlying network infrastructure faces challenges from addressing maintenance, security, performance, and scalability to make the network more reliable and stable. Software-defined networking, blockchain, and network function virtualization were proposed and realized to address such issues in both academic and industry wise. This paper analyzes and summarizes works from implementing different categories of blockchains as an element or enabler of network functions to resolve the limitation. Blockchain as a network function has been proposed to give support to the underlying network infrastructure to provide services that have less lag, are more cost-effective, have better performance, guarantee security between participating parties, and protect the privacy of the users. This paper provides a review of recent work that makes use of blockchain to address such networking related challenges and the possible setbacks in the proposal.
Blockchain, Performance, Security, Network Function
Jamolbek Mattiev1 and Branko Kavšek1,2, 1Department of Information Science and Technologies, University of Primorska, Koper, Slovenia, 2Artificial Intelligence Laboratory, Jozef Stefan Institute, Ljubljana, Slovenia
Existing well-known classification rule based algorithms use mainly greedy heuristic search to find regularities in datasets for classification. In recent years, extensive research on association rule mining was performed in the machine learning community on learning rules by using exhaustive search. The main aim of association rule mining is to find all rules in datasets that satisfy the user-specified minimum support and minimum confidence thresholds. Even though the all rules may not be used directly for effective classification, accurate and efficient classifiers have been built using these rules, so called, classification association rules. In this paper, we compare “classical” classification rule learning algorithms that uses greedy heuristic search with a class association rule(car) learner that uses constrained exhaustive search to find classification rules on real-life datasets. We propose a simple method to extract the reasonable number of class association rules by grouping them to form an accurate classifier. We have performed experiments on 12 datasets from UCI Machine Learning Database Repository and compared the results with well-known rule based and tree based classification algorithms. Experimental results show that our method was consistent and comparative with others, our proposed method achieves the fourth highest result among all classification algorithms (rule based and tree based although) on average accuracy with 82.16%. Although Random Forest, Ripper and PART algorithms gained the highest results on average accuracy, our method achieved the best accuracy (81%) on “Breast Cancer” dataset.
Attribute, frequent Itemset, Minimum Support, Minimum, Confidence, Class Association Rules, Classification.
Zeinab FARHAT1, Ali KAROUNI2, Bassam DAYA3 and Pierre CHAUVET4 1EDST,Lebanese University,Lebanon,Beirut,2,3University Institute of Technology,Lebanese University,Lebanon,Sidon,4LARIS EA, Angers University France, France, Angers
Nowadays, road traffic accidents are one of the leading causes of deaths in this world. It is a complex phenomenon leaving a significant negative impact on human’s life and properties. Classification techniques of data mining are found efficient to deal with such phenomena. After collecting data from Lebanese Internal Security Forces, data are split into training and testing sets using 10-fold cross validation. This paper aims to apply two different algorithms of Decision Trees C4.5 and CART, and various Artificial Neural Networks (MLP) in order to predict the fatality of road accidents in Lebanon. Afterwards, a comparative study is made to find the best performing algorithm. The results have shown that MLP with 2 hidden layers and 42 neurons in each layer is the best algorithm with accuracy rate of prediction (94.6%) and area under curve (AUC 95.71%).
Data mining, Fatal Road Accident Prediction, Neural Networks, Decision trees.
Nicholas Patterson, Deakin University, Australia
Virtual worlds have become highly popular in recent years with reports of over a billion people accessing these environments and the virtual goods market growing to near 50 billion US dollars. An undesirable outcome to this popularity and market value is thriving criminal activity in these worlds. The most profitable cyber security problem in virtual worlds is named Virtual Property Theft. The aim of this study is to use an online survey to gain insight into how hackers (n=100) in these synthetic worlds conduct their criminal activity. This survey is the first to report an insight into the criminal mind of hackers (virtual thieves). Results showed a clear-cut profile of a virtual property thief, they appear to be mainly aged 20- 24 years of age, live in the United States of America, while using virtual worlds for 5-7 hours a day. These and the other key results of this study will provide a pathway for designing an effective anti-theft framework capable of abolishing this cyber security issue.
Virtual world environments, virtual property theft, cyber security, hackers, massively multiplayer online games
Natarajan Meghanathan,Jackson State University, 1400 Lynch St, Jackson, MS, USA
Until now, we were determining stable data gathering (DG) trees for mobile sensor networks (MSNs) using a link stability metric (computationally-light or computationally-heavy) that is directly computed on the egocentric edge network. Among such DG trees, the BPI' (complement of bipartivity index)-based DG trees were observed to be the most stable, but the BPI' metric is also computationally-heavy. Hence, we seek to build a multi-variable linear regression model to predict the BPI' values for the egocentric networks of edges using three computationally-light metrics (neighborhood overlap: NOVER, one-hop two-hop neighborhood: OTH, and normalized neighbor degree: NND) that are also computed on the egocentric edge networks. The training and testing are conducted as part of a single simulation run (i.e., in-situ). The training dataset comprises of the BPI', NOVER, OTH and NND values of randomly sampled egocentric edge networks during the first phase of the simulation (1/5th of the total simulation time). We observe the R-square values for the prediction to be above 0.85 for both low density and high density networks. We also observe the lifetimes of the predicted BPI'-based DG trees to be 87-92% and 55-75% of the actual BPI'-based DG trees for low-moderate and moderate-high density networks respectively.
Multi-variable Regression, Bipartivity Index, Computationally-Light, Computationally-Heavy, Mobile Sensor Networks, Data Gathering Tree