ICADS '23 Keynote: AI-based Decision Frameworks for Smart Environments – Some Case Studies

ICADS ’23: Second International Conference on Applied Data Science For access to past video webinars or to join our Dlist to hear about future programs, please visit https://r6.ieee.org/scv-cs

The 1st case study is Online Examination Question Classification Systems. The advancement in education has emphasized the need to develop an efficient online examination system. Here, the assessment of the exam questions carried out qualitatively and the cognitive levels of students are mapped to the Bloom’s cognitive levels. This helps evaluate the subject-related learning by the students. For this task, a novel optimized framework, referred to as QC-DcCapsGAN-AOSA is developed by combining the Dual-channel Capsule Generative Adversarial Network (DcCapsGAN) with the Atomic Orbital Search Algorithm (AOSA) for preprocessing a real-time online exam questions dataset of Indian universities. Key features from the raw data are identified using Term Frequency Inverse Document Frequency (TF-IDF) and eventually classifying the exam questions. AOSA is used to fine-tune the parameters’ weights of the DcCapsGAN, and then these weights are used to categorize the exam questions as Knowledge Level, Comprehension Level, Application Level, Analysis Level, Synthesis Level, and Evaluation Level. Experimental results demonstrate the superiority of the QC-DcCapsGAN-AOSA framework when compared to the state-of-the-art methods such as QC-LSTM-CNN and QC-BiGRU-CNN. The 2nd case study is on Online Exam Proctoring Systems, which have become increasingly important in education sector due to the need for remote learning during the COVID-19 pandemic. However, ensuring the integrity of online exams is a major challenge. Traditional proctoring methods may not be suitable for remote learning scenarios. To address this challenge, Prof. Reddy’s team proposed an online proctoring system that uses deep learning and federated learning techniques to detect fraudulent behaviors during online exams. In this work, they use four transfer learning models, namely: VGG16, ResNet18, MobileNetV2, and DenseNet121, for classifying images captured during the exam to identify cheating behaviors on created online exam dataset. They train on this dataset using a federated learning approach to ensure the privacy of the students. The performance of both the centralized and decentralized (federated) models with transfer learning is carried out in terms of accuracy, recall, precision and F1-Score metrics. The results demonstrate that the federated learning with DenseNet121 transfer learning model is superior to all other considered models. The 3rd use case is Smart Home Kitchen Appliance, specifically, a Smart Refrigerator. A smart kitchen in a smart home homes IoT and AI-enabled appliances and devices that assure comfort, convenience, safety and protection to its users. Most smart appliances available in the market are not economical and are beyond the reach of a commoner. A Smart fridge is one such appliance. Considering this, we propose a cost-effective, ubiquitous, and intelligent refrigerator framework. The proposed framework uses Night Vision images to detect and predict fridge items and provides a natural language interaction feature with the fridge. The framework aims to convert any standard refrigerator into its smarter version with minimal hardware and software requirements. The design allows users to view fridge contents on the go using a mobile application and interact with it using natural language. A YOLOv5 model and a custom dataset of Night Vision images are used for object detection. The object detection model achieved mAP of 97.1% compared to YOLOv3-tiny and YOLOv4-tiny models, whose mAPs are 94.8% and 96.3%, respectively. For access to past video webinars or to join our Dlist to hear about future programs, please visit https://r6.ieee.org/scv-cs