In aquaculture, feeding cost usually takes a lot from the total budget. However, a significant portion of the feed often gets uneaten by the fish. To overcome this problem, many ways have been developed using artificial intelligence to detect fish's appetite levels. In this paper, a deep learning model using multi-modal sensors is implemented to detect fish's appetite level based on its visual and the aquarium's water movement. We propose the combination R(2+1)D-CNN and 1D-ANN to process the sensor data, in the form of video streams and accelerometer signals, respectively. Based on the result, this model significantly increases the accuracy of the inference (from 86.71% to 91.84% testing accuracy) without adding many learning parameters (from 1,777,186 to 1,833,008 parameters), compared to the model using a single-modal deep learning model with the camera.
Parents' commitment and engagement are crucial elements in collaborative online learning (COL). The continuous support provided by teachers and schools can foster parent-teacher engagement within the school community. Such engagement from parents yields benefits for both the school and students, leading to improved performance and academic achievements. However, several barriers hinder parent-teacher engagement, including parental commitments, time constraints, limited digital literacy, technical and communication skills, and diverse cultural backgrounds. To address these challenges, technology and various communication methods, such as social media and online learning platforms, have been recognised as potential solutions to enhance parent-teacher engagement. Therefore, this study aims to overcome these barriers and provide solutions by introducing the COL framework for cultivating parent-teacher engagement at primary school. This study reveals this framework's effectiveness in promoting continuous parent-teacher engagement.
The internet and computer applications are useful to everyday life but without proper control, it can threaten everyone's life. Now, the dependence on the use of internet and social media has opened the risk to the new threat which is cybercrime. Cybercrime is a crime through the use of computers, where it has become a major challenge faced globally. Therefore they (especially students) need to be exposed to knowledge about cybercrimes such as cyber bullying, cyber stalking, cyber harassment, identity theft, and others. Awareness of this crime is necessary so that they can provide preventive measures by controlling themselves from getting caught up in this crime. Therefore, a cyber education module that is compatible with Malaysians' culture needs to be developed. The importance of cyber education among school students has attracted the interest of many parties. However, the educational approach is geared towards traditional methods through lectures or informal learning. Therefore, the cyber learning package is necessary to give exposure to students in increasing their awareness of the cybercrime threats. Based on reviews on the development of cyber education, until today there has been no comprehensive effort in developing interactive cyber learning package that are suitable for Malaysians' culture. This study used STEM (Science, Technology, Engineering and Mathematics) learning approach to increase student motivation in cyber education. To support STEM approach, five learning principles of 5E (namely engage, explore, explain, elaborate, and evaluate) is adopted. The advantage of using this learning principle is it encourages and motivates students in learning as well as give proper guidance in skill development. To adopt STEM approach, a learning model is proposed that consists of 5E learning principles in cyber education.
Computer vision applications have been widely used in critical areas however its reliability is hampered by low accuracy when dealing with small objects. As the internal information for small objects is limited by its pixels, addition of external information from the object context is important to determine object identity. Context features that are shared among objects can be used to group and label them hierarchically. It will then be used to improve classification by reducing the search space and linking the object's general feature or function. Hierarchical-based CNN uses a hierarchy of labels with multi-level classifiers in classification of simpler coarser level before tackling more complex classification on finer level. However, the current implementation of hierarchical CNN keeps information between levels isolated from each other thus limiting the potential of its structure. Improvement is made towards the loss function with the addition of pathway loss where the hierarchy pathway from coarse to fine is considered on each classifier branch. Additionally, we introduce the reference forwarding function, which transfers crucial information learned from previous layers to the subsequent finer ones, thereby further improving the classification accuracy. Proof of concept experiments on CIFAR-100 dataset shows each of the proposed innovations improve accuracy compared to baseline model and the overall combinations synergize well with each other, achieving the best result with improvement of almost 8%.
Diabetes mellitus, particularly type-2 diabetes, remains a prevalent health issue, raising concerns due to its associated risk of complications. Among these, cardiovascular complications pose a significant threat, exhibiting high morbidity and mortality rates. Health screening plays a pivotal role in stratifying the risk levels of diabetes patients, facilitating proactive measures to prevent the progression of complications. As such, the primary objective of this study is to develop a predictive model system for assessing cardiovascular risk in diabetes patients. Our study used the Cardiovascular Disease dataset and conducts experiments with various supervised machine learning algorithms, such as Naive Bayes, decision tree, random forest, AdaBoost, and XGBoost. The results reveal that ensemble learning algorithms based on boosting, particularly AdaBoost and XGBoost, outperform other supervised machine learning methods. However, even with the best performance achieved using the dataset, the accuracy stands at 0.71, and the F-1 score is 0.69, which is still acceptable for screening purposes. Although these results provide valuable insights, indicating individuals at higher risk for cardiovascular complications in diabetes, further improvements are needed to enhance early prevention strategies.
