The 6th International Conference on Wireless, Intelligent and Distributed Environment for Communication

There is an increased number of vehicles on university campuses such as the University of Technology–Jamaica. This necessitates better tracking of vehicles entering and leaving the campus and identifying parking violations. One of the more prominent ways of addressing these issues in other organizations is the implementation of automated parking systems. These implementations generally use sensor technology for vehicle detection along with artificial intelligence in the form of automatic license plate readers to recognize and record vehicle license plates. Automated parking systems, however, generally lack features that are specifically designed to aid in security and are not often used by institutions and businesses to aid in management. We designed a system, which we named Campus Vision, to tackle misplacement, tracking, and parking of vehicles on the University of Technology, Jamaica, Papine Campus. This chapter presents our methodology, results, conclusions, and recommendations. We designed our system using the python programming language along with the YOLO and EasyOCR libraries. The system analyzed, extracted, and stored images from the live stream of video from select security cameras at the university. Our results showed that Campus Vision successfully identified 78% of vehicles entering and 71% of vehicles exiting the campus, and of these vehicles identified, it recognized 98% of the license plates. Campus Vision successfully detected 90% of submitted vehicles involved in parking violations, with 100% license plate detection. We concluded that Campus Vision was able to achieve an overall accuracy rate of 76% in detecting vehicles and 98% in detecting license plates while only being able to achieve a 34% accuracy rate in correctly identifying license plate numbers. We recommend that the system be considered for implementation but with dedicated GPUs and better strategic placement of security cameras. We further recommend that more research be conducted in this area, for example, with motorcycles, trucks, and buses on campus.

In an era where cybercrime is getting more and more complex by day, the accuracy in classifying credit card transactions is an important issue. This chapter argues if the weighted voting stacking ensemble method that combines various classifier models can be utilized as solution to this issue. But in using such solution, selecting the appropriate weights of classifier models for the correct classification of credit card transactions is a problem, which can be viewed as a problem of weights optimization. This chapter proposes to use the differential evolution optimization method as approach to define the appropriate weight function for the weight voting stacking ensemble method of various classification methods. It is found that the number of true positive transactions generated by the stacking ensemble method is exceeded (0.7% vs 85.4%) by that of the true positives generated using the differential evolution optimization.

Advances in the development of electronic components, particularly microprocessors, have enabled the manufacture of increasingly small and intelligent devices such as wireless sensors. These sensors, among other uses, are networked in a special type of ad hoc network called a wireless sensor network (WSN) and used in a variety of real-world applications, such as environmental monitoring, smart healthcare, or, more generally, where it is desired to be able to monitor a remote environment. Secure communication is an essential element of these types of networks because it requires the implementation of special mechanisms to ensure an acceptable level of security given the limited resources of the network equipment (sensors). In this chapter, we propose a scheme based on the functioning of the A-star algorithm to create paths and thus keys between the nodes of a network. The scheme exploits an algorithm using a cost function based on signal strength and cryptographic resources (cryptographic computations and memory occupied by the keys). The proposed scheme introduces a method of pre-distribution of keys with periodic renewal as well as on-demand renewal while simplifying the key management. Intensive simulations show that the proposed method outperforms the LEAP+ approach by around 13.21% in terms of RAM usage.

Vehicular edge networks are pivotal in delivering services and applications that rely heavily on efficient resource allocation. Various strategies utilize intelligence, prediction, optimization, and incentive modelling to ensure optimal functioning within these networks. Despite the advancements, vehicular networks face persistent challenges that inhibit efficient resource allocation and communication. The most notable challenges are sporadic connectivity, transmission delays, and inherent uncertainty due to highly dynamic environments. In light of these challenges, integrating graph neural networks (GNNs), which learn hidden spatial and functional patterns of complex vehicular networks with clustering methodologies, emerges as a promising solution. By harnessing the power of GNNs and clustering, this approach provides an opportunity for more intelligent organization of the network nodes to reduce transmission delays and to improve resource allocation in highly dynamic environments. It creates a holistic environment that supports predictions and estimates based on trending communication and mobility features.

Optimization of energy consumption in Internet of Things (IoT) Wireless Network and routing of payload data are major concerns. In this chapter, we combine several machine learning algorithms to address these issues. The proposed sequential approach is based on two algorithms. First, we put forward the Density-Based Spatial Clustering of Applications with Noise (DBSCAN), in which IoT nodes are grouped into clusters. Second, we built an intra-cluster prediction model based on multiple linear regression, in which the cluster head (CH) predicts the next information of each cluster member. This helps to limit the communication between CH and members and thus reduce the network energy consumption. Third, we performed CH selection according to its residual energy, the distance from the base station, and the base station and the number of cluster members. Simulations and a comparative study have been carried out to prove the relevance of the proposed method. From experimental analyses, we found that the proposed method increases network lifetime by a factor of 13.16, 28.75, and 47.66 as compared to LEACH, k-means-SDR, and DBSCAN-SDR approaches, respectively.

Opportunistic network (OppNet) is a subset of Mobile Ad hoc Networks (MANETs) and a kind of Delay Tolerant Network (DTN), formed by wirelessly connected nodes that are mobile in nature. The routes established by such networks for data transmission is dynamic in nature. This leads to challenges for researchers to develop routing protocols to transmit messages in OppNet. This work proposes a novel routing protocol for forwarding messages from source to destination in OppNet, so-called location-based clustering approach for next-hop selection in opportunistic networks (LBC). In this proposed work, several cluster points are identified within the network where more nodes generally gather. During message transmission, distance between the cluster of the neighbour node of sender and that of destination is calculated. If this distance lies within a specified threshold, packet is forwarded to that node. The proposed model is compared with existing work—Epidemic, Prophet, Encounter and Distance-based Routing (EDR) and Encounter Buffer and Contact duration-based Routing (EBC) protocols, to name a few. It is found from simulations that in terms of average latency, LBC outperforms Epidemic, Prophet, EDR and EBC, respectively, by 12.46%, 10.63%, 35.9% and 15.38%, when the number of nodes is increased.

This chapter studies the usage of artificial intelligence in network analysis when applied to the problem of attempting to identify security vulnerabilities. This chapter demonstrates how AI is capable of more efficient and detailed analysis than a human cyber-security individual by using the CyberBattleSim simulation program developed by Microsoft, which utilizes Deep-Q-learning algorithms to train the AI. During the investigation conducted in this report, several environments are studied: three networks of varying size based on the chain network pattern provided by the initial program, three networks based on the toy network pattern, and a custom network designed to examine the AI’s capabilities in handling a unique environment involving honeypots and two-factor authentication. The AI was determined to be adept at learning from one episode to the next, accumulating rewards more quickly, and using less time to take control of the network. Based on the research conducted, multiple modifications to improve the system are suggested.

Heart failure is a chronic medical condition characterized by impaired contractile properties of the heart muscles, resulting in reduced blood supply within the body. According to the World Health Organization, heart failure is one of the most serious ailments that limit the patient’s activities and decreases their life span, contributing to the highest number of deaths globally. Highlighting the significance of risky factors related to the survival of patients with heart failure is crucial for effective intervention. In this chapter, we propose a new multifactorial approach to explain risk features for predicting the survival rate of heart failure using a multivariate logistic regression model. The experimental evaluation with a medical dataset collected from 299 patients diagnosed with heart failure shows that the most risk factors include age, ejection fraction rate, and serum creatinine level in terms of binary categories of patients with smoking, diabetes, high blood pressure, and anemia. In particular, the likelihood of patients experiencing a fatal event following heart failure will increase by 3.0 for each additional year of age. When the ejection fraction increases by 1%, the odds of patients experiencing a fatal event decrease by 2.66, and a one-unit rise in serum creatinine results in the odds of patients experiencing a fatal event following heart failure increasing by a factor of 0.41.

Parameter specification remains a difficult task in data stream clustering as density-based algorithms hyperparameters tuning to their optimal values are often difficult to determine. This paper investigates the sensitivity of parameter tuning on DenStream, a data stream clustering algorithm. The effects on different noise levels are evaluated for the DenStream and two chosen benchmark algorithms, CluStream and ClusTree algorithms, using both synthetic and real-world datasets, and several performance metrics. It was found that DenStream outperforms CluStream and ClusTree on some of those metrics.

The healthcare industry has witnessed a significant transformation due to the emergence of open-source medical cyber-physical systems, primarily driven by advancements in 3D printing technology. However, the growing use of these open-source systems in hospitals has also brought about cybersecurity concerns. In particular, adopting new technologies, such as mobile medical devices, has introduced new challenges when handling distributed denial-of-service (DDoS) attacks. Despite numerous statistical methods developed for DDoS attack detection, there remains a prominent concern regarding developing real-time detectors with low computational overhead. In addition, evaluating new detection algorithms and techniques heavily relies on the availability of well-designed datasets. Therefore, we create in this chapter a new dataset called MedibotDDoS, including zero-day DDoS attacks, and perform a comparative study involving various machine learning algorithms based on a collection of network flow features utilizing the generated dataset to develop an effective strategy for detecting such DDoS attacks. The experimental evaluation results show that the random forest classifier performs the best and achieves the highest overall accuracy of 99.998%.

The surge in mobile device adoption has led to an increase in cyber threats, particularly Short Message Service (SMS) phishing (smishing), which poses significant risks to personal and financial security. To address this growing concern in the context of Ghana, we have introduced Smishield—a machine-learning-based smishing detection system. By utilizing Term Frequency-Inverse Document Frequency (TF-IDF) vectorization, SMS messages are converted into numerical representations, effectively capturing their underlying meaning. Our evaluation encompasses a variety of classification algorithms, such as Support Vector Machines, Logistic Regression, Decision Trees, and Random Forest, while considering different test and training ratios. Through thorough assessment, the Random Forest Tree algorithm emerges as the optimal candidate, boasting remarkable statistics of 99.47% accuracy, 98.91% precision, and an impressive Area Under the Curve (AUC) score of 99.95%.

In the context of our study, the model derived from our efforts is named SmishShield. We conducted a comprehensive evaluation by benchmarking it against other models, and the results demonstrate its notable performance.