Scalable and secure authorization of smart things is of the crucial essence for the successful deployment of the Internet of Things (IoT). Unauthorized access to smart things could exacerbate the security and privacy concern, which could, in turn, lead to the reduced adoption of the IoT, and ultimately to the emergence of severe threats. Even though there are a variety of IoT solutions for secure authorization, authorization schemes in highly dynamic distributed environments remain a daunting challenge. Access rights can dynamically change due to the heterogeneous nature of shared IoT devices and, thus, the identity and access control management are challenging. This survey provides a comprehensive comparative analysis of the current state-of-the-art IoT authorization schemes to highlight their strengths and weaknesses. Then, it defines the most important requirements and highlights the authorization threats and weaknesses impacting authorization in the IoT. Finally, the survey presents the ongoing open authorization challenges and provides recommendations for future research.

Crime Pattern Theory is a way of elucidating the reasons why specific types of crime happen at certain areas only. According to the theory, the offenders, rather than venturing into unknown territories, frequently commit crimes by taking advantage of the opportunities they encounter in the places of their comfort zones or the places they are most familiar with. Spatial analysis of crimes and areas of higher concentration assist in preventing or at least reducing the amount of future crime from the knowledge discovery of past data. Our approach of crime prediction consists of five modules, namely data extraction, pre-processing, classification, pattern identification, prediction and google map visualization. This paper compares machine learning algorithms namely naive bayes, bayesian networks, weighted k-nearest neighbors, multi-layer perceptron classifier, guassian naive bayes, decision tree, random forest, adaboost, gradient boosting, linear discriminant analysis and quadratic discriminant analysis for crime category identification and crime prediction. The results have been evaluated on a real dataset of crimes in the city of San Francisco. The work successfully uses the temporal and spatial information in the data to locate the crime hotspots, predict the category of a crime in a region and additionally, the emergence of the crimes.

Hearing loss is a disease exhibiting a growing trend due to a number of factors, including but not limited to the mundane exposure to the noise and ever-increasing size of the older population. In the framework of a public health policymaking process, modeling of the hearing loss disease based on data is a key factor in alleviating the issues related to the disease and in issuing effective public health policies. First, the paper describes the steps of the data-driven policymaking process. Afterward, a scenario along with the part of the proposed platform responsible for supporting policymaking are presented. With the aim of demonstrating the capabilities and usability of the platform for the policy-makers, some initial results of preliminary analytics are presented in the framework of a policy-making process. Ultimately, the utility of the approach is validated throughout the results of the survey which was presented to the health system policy-makers involved in the policy development process in Croatia.

The advent of the COVID-19 pandemic has brought several never-before-seen changes in the daily lives of people around the globe. As a way to curb the spreading of the disease, wearing face masks has become mandatory in the majority of public places. To solve the necessity of face mask detection in such situations, there have been only a handful of research endeavors up to this date. Computer vision has advanced multi-fold with the advent of AlexNet architecture. With a motivation to go deeper with the neural network architecture, the concept of Depthwise Separable Convolutions and projection layer was developed in MobileNetV1. In this work, a novel lightweight deep learning model based on Single Shot Detector (SSD) MobileNetV2 architecture is proposed for face mask detection using images and video streams of crowds aiming its utilization on low compute re-source environment. An open benchmark face mask dataset, with 4095 images including masked and no mask images, is utilized to train the model for detection. The model is initialized using transfer learning with the freezing of base layers. The proposed methodology can efficiently aid in tracking and enforcing social distancing rules in crowded places with the use of surveillance cameras. On the different benchmarks that we have tested, the model proved to be highly successful and has achieved an accuracy rate of 99.39% and an F1 score of 0.995.