The applications of machine learning and deep learning in demand forecasting have attracted increasing attention, as they offer remarkable predictive capabilities that help automate forecasting processes and achieve higher accuracy. While numerous review studies have examined solutions within specific industries, there is a lack of comprehensive literature review investigating these solutions across different sectors. Therefore, this study overviews machine learning and deep learning applications in demand forecasting across time-critical industries, including power, tourism, water, transportation, and food. A two-tier classification framework is proposed to categorize demand forecasting studies by both application industry and methodological architecture. In addition, the most popular statistical metrics for evaluating demand forecasting are reviewed and summarized. This study reveals that while machine learning and deep learning are effective for demand forecasting, model selection highly depends on the target industry, data availability, and computational resources. Therefore, this study proposes a conceptual, generic framework that maps data characteristics to appropriate model architecture classes for demand forecasting and recommends adopting scale-independent evaluation metrics. The proposed framework offers a structured pipeline and practical guidance for practitioners and researchers to design forecasting systems across diverse industries, enabling consistent comparative analysis.

The convergence of urban informatics and vehicle intelligence has given rise to smart connected vehicles, which have immense potential as edge computing platforms for various applications. However, harnessing the full efficiency of these platforms presents challenges due to the diverse resource requirements, capabilities, and vehicle types, as well as unpredictable vehicle movements. To address these obstacles, a novel task offloading framework based on digital twin (DT) technology has been proposed for the Internet of Vehicles (IoV). This DT-based framework capitalizes on historical data and workload predictions to optimize the utilization of edge devices. It streamlines the offloading process by enabling tasks to be accepted and processed by the source vehicle without relying on external devices. The proposed system is designed to learn and forecast vehicle mobility patterns and computation waiting times, facilitating efficient allocation of computing resources at edge locations. Consequently, this approach enhances the quality of service by ensuring swift and effective task processing, irrespective of the vehicles’ unpredictable movements. The proposed approach is compared with a deep sequential model based on reinforcement learning, collaborative multiaccess edge computing (MEC), and energy-efficient MEC via reinforcement learning model. Our method demonstrates an improvement in task execution and overall offloading performance compared to these techniques during peak vehicle arrival rates. Likewise, substantial enhancements are observed in other benchmark parameters.

Abstract Social bots are automated social media accounts governed by software and controlled by humans at the backend. Some bots have good purposes, such as automatically posting information about news and even to provide help during emergencies. Nevertheless, bots have also been used for malicious purposes, such as for posting fake news or rumour spreading or manipulating political campaigns. There are existing mechanisms that allow for detection and removal of malicious bots automatically. However, the bot landscape changes as the bot creators use more sophisticated methods to avoid being detected. Therefore, new mechanisms for discerning between legitimate and bot accounts are much needed. Over the past few years, a few review studies contributed to the social media bot detection research by presenting a comprehensive survey on various detection methods including cutting-edge solutions like machine learning (ML)/deep learning (DL) techniques. This paper, to the best of our knowledge, is the first one to only highlight the DL techniques and compare the motivation/effectiveness of these techniques among themselves and over other methods, especially the traditional ML ones. We present here a refined taxonomy of the features used in DL studies and details about the associated pre-processing strategies required to make suitable training data for a DL model. We summarize the gaps addressed by the review papers that mentioned about DL/ML studies to provide future directions in this field. Overall, DL techniques turn out to be computation and time efficient techniques for social bot detection with better or compatible performance as traditional ML techniques.

Facial emotion detection and expressions are vital for applications that require credibility assessment, evaluating truthfulness, and detection of deception. However, most of the research reveal low accuracy in emotion detection mainly due to the low quality of images under consideration. Conducting intensive pre-processing activities and using artificial intelligence especially deep learning techniques are increasing accuracy in computational predictions. Our research focuses on emotion detection using deep learning techniques and combined preprocessing activities. We propose a solution that applies and compares four deep learning models for image pre-processing with the main objective to improve emotion recognition accuracy. Our methodology includes three major stages in the data value chain, pre-processing, deep learning and post-processing. We evaluate the proposed scheme on a real facial data set, namely Facial Image Data of Indian Film Stars for our study. The experimentation compares the performance of various deep learning techniques on the facial image data and confirms that our approach enhanced significantly the image quality using intensive pre-processing and deep-learning, improves accuracy in emotion prediction.

Cloud services are widely used to virtualize the management and actuation of the real-world the Internet of Things (IoT). Due to the increasing privacy concerns regarding querying untrusted cloud servers, query anonymity has become a critical issue to all the stakeholders which are related to assessment of the dependability and security of the IoT system. The paper presents our study on the problem of query receiver-anonymity in the cloud-based IoT system, where the trade-off between the offered query-anonymity and the incurred communication is considered. The paper will investigate whether the accepted worst-case communication cost is sufficient to achieve a specific query anonymity or not. By way of extensive theoretical analysis, it shows that the bounds of worst-case communication cost is quadratically increased as the offered level of anonymity is increased, and they are quadratic in the network diameter for the opposite range. Extensive simulation is conducted to verify the analytical assertions.

The data in higher educational institutions come from the interaction of students with the various online systems, such as learning management, registration, advising and email. Research in the field of educational data mining is concerned with the collection and analysis of such data to discover new insights about student behavior, learning style and success factors. Since different departments in an institution manage different IT systems, collecting data from all of these departments requires collaboration. The data has to be extracted from many systems, which uses different data formats. Therefore, a typical research work in this field analyzes data extracted from one system. This paper, on the other hand, analyzes the network traffic that students generate while on-campus. This approach provides us with a better view of the student interaction with the educational systems, compared to the single view achieved by analyzing data from one system. We anonymize student personally identifiable information to protect student privacy. Further, we propose the use of fog computing to enhance student privacy and reduce network load.

Little research effort contributed to help learners reflect on their current cognitive skills. A storehouse of this valuable information could provide valuable indications to experienced mentors to intervene and guide learners improving their learning capabilities. Learning how to learn using the proposed cognitive apprenticeship approach in this paper goes beyond the given classroom knowledge to reflect on how to maximize learning readiness under the supervision of experienced mentors. In this paper, we promote an autonomous inquiry-based learning cycle to support formal education processes, whereby students learn from individual discoveries and from peers. At the completion of each inquiry-based learning cycle, a cognitive apprenticeship cycle is triggered to assess the improvements in learning dispositions and guide apprentices in developing their own cognitive skills. We propose an online instrument for collecting self-assessment data and communicating with mentors, to produce willing, confident and creative lifelong learners. The provided instrument presents a storehouse view of learning disposition through an integrated portal which captures self-stated learning experiences and converts them into analytical results to diagnose deficiencies and prescribe improvement recommendations. In doing so, we provide "learning disposition" dimensions along which, we analytically measure learning aptitudes of individual learners. The proposed tool in this project generates an inventory of learning capabilities to guide learners reorganizing their learning agenda. Collected data out of the proposed tool reflects learners' perception at a given point of time. The tool is applied again at a later point of time to assess the changes and move the apprentice to the next level of his cognitive skills.