Federated learning (FL) provides a distributed machine learning system that enables participants to train using local data to create a shared model by eliminating the requirement of data sharing. In healthcare systems, FL allows Medical Internet of Things (MIoT) devices and electronic health records (EHRs) to be trained locally without sending patients data to the central server. This allows healthcare decisions and diagnoses based on datasets from all participants, as well as streamlining other healthcare processes. In terms of user data privacy, this technology allows collaborative training without the need of sharing the local data with the central server. However, there are privacy challenges in FL arising from the fact that the model updates are shared between the client and the server which can be used for re-generating the client’s data, breaching privacy requirements of applications in domains like healthcare. In this paper, we have conducted a review of the literature to analyse the existing privacy and security enhancement methods proposed for FL in healthcare systems. It has been identified that the research in the domain focuses on seven techniques: Differential Privacy, Homomorphic Encryption, Blockchain, Hierarchical Approaches, Peer to Peer Sharing, Intelligence on the Edge Device, and Mixed, Hybrid and Miscellaneous Approaches. The strengths, limitations, and trade-offs of each technique were discussed, and the possible future for these seven privacy enhancement techniques for healthcare FL systems was identified.

Cybersecurity is one of the important considerations when adopting IoT devices in smart applications. Even though a huge volume of data is available, data related to attacks are generally in a significantly smaller proportion. Although machine learning models have been successfully applied for detecting security attacks on smart applications, their performance is affected by the problem of such data imbalance. In this case, the prediction model is preferable to the majority class, while the performance for predicting the minority class is poor. To address such problems, we apply two oversampling techniques and two undersampling techniques to balance the data in different categories. To verify their performance, five machine learning models, namely the decision tree, multi-layer perception, random forest, XGBoost, and CatBoost, are used in the experiments based on the grid search with 10-fold cross-validation for parameter tuning. The results show that both the oversampling and undersampling techniques can improve the performance of the prediction models used. Based on the results, the XGBoost model based on the SMOTE has the best performance in terms of accuracy at 75%, weighted average precision at 82%, weighted average recall at 75%, weighted average F1 score at 78%, and Matthews correlation coefficient at 72%. This indicates that this oversampling technique is effective for multi-attack prediction under a data imbalance scenario.

Solving the university timetabling problem is crucial as it ensures efficient use of resources, minimises scheduling conflicts, and enhances overall productivity. This paper presents a comprehensive review of university timetabling problems using integer programming algorithms. This study explores various integer programming techniques and their effectiveness in optimising complex scheduling requirements in higher education institutions. We analysed 95 integer programming-based models developed for solving university timetabling problems, covering relevant research from 1990 to 2023. The goal is to provide insights into the evolution of these algorithms and their impact on improving university scheduling. We identify that the implementation rate of models using integer programming is 98%, which is much higher than 34% implementation rates using meta-heuristics algorithms from the existing review. The integer programming models are analysed by the problem types, solutions, tools, and datasets. For three types of timetabling problems including course timetabling, class timetabling, and exam timetabling, we dive deeper into the commercial solvers CPLEX (47), Gurobi (11), Lingo (5), Open Solver (4), C++ GLPK (4), AIMMS (2), GAMS (2), XPRESS (2), CELCAT (1), AMPL (1), and Google OR-Tools CP-SAT (1) and identify that CPLEX is the most frequently used integer programming solver. We explored the uses of machine learning algorithms and the hybrid solutions of combining the integer programming and machine learning algorithms in higher education timetabling solutions. We also identify areas for future work, which includes an emphasis on using integer programming algorithms in other industrial areas, and using machine learning models for university timetabling to allow data-driven solutions.

Government funding programs administered by the Australian Research Council (ARC) aim to advance national research priorities while generating scholarly and socio-economic impact. This study employs a descriptive bibliometric benchmarking approach to examine the relationship between funding levels and scholarly output for publications explicitly acknowledging ARC support. Using project-level funding data linked with journal articles published between 2009 and 2016, we analyze 10,565 ARC-funded projects receiving a total of AUD 4.6 billion and producing 54,639 journal publications. On average, each project received approximately AUD 437,720 and generated five publications, corresponding to a cost of about AUD 84,700 per article. We compare research productivity, citation impact, and return on investment across ARC Discovery and Linkage programs, as well as between STEM and HASS disciplines. The results reveal no strong correlation between funding amount and either publication volume or citation impact across ARC programs. STEM projects generally exhibit higher returns on investment and citation impact; however, a subset of HASS projects achieves exceptionally high efficiency relative to funding received. Notably, projects funded below AUD 100,000 demonstrate the highest return on investment in terms of both publication productivity and normalized citation impact. These findings suggest that smaller grants can yield disproportionately high scholarly returns, offering important implications for research funding allocation, efficiency evaluation, and performance assessment in public research systems.

(1) Background: Evaluating popular educational chatbots (EC) shows that existing EC impacts users' ability to synthesize and take ownership of their work, potentially leading to misinformation. This study aims to design and evaluate a research-guided chatbot that improves research synthesis. (2) Method: One hundred sixty-five graduate students evaluated the chatbot. The qualitative data were analysed using natural language processing for sentiment analysis and emotion detection. (3) Results: The findings show that "interesting" and "enjoyable" were frequent words used in user feedback. Sentiment analysis findings show that 47% suggested users are satisfied with the chatbot's recommendation potential; 46.3% found the guided chatbot to be the most effective in simplifying research methodology concepts; 48.4% identified the chatbot as the most effective tool for conceptualizing research questions and hypotheses and excellent in fostering ownership of their ideas; 85.3% rated the chatbot as fostering creativity; and 81.0% rated the chatbot positively for its ability to minimize misinformation. Two themes were identified in the thematic analysis: attitudes and improvements to future research-guided chatbots, emphasizing simplicity. (4) Conclusions: The strong preference for the guided chatbot underlines the potential for future improvements in EC with the flexibility to gain ownership of work with a guided approach.

Accurate prediction of intrusion attacks is crucial in the era of Internet of Things to detect and respond promptly to suspicious activities to minimise the damage caused by those attacks. The instances of attacks are always much less than normal traffic and machine learning models are used to detect those comparatively smaller attack classes. For machine learning models the imbalance of classes in the training data results in better accuracy of prediction for majority classes and much less accurate detection of minority classes. In this paper, we propose a class-based classification framework to address the imbalance problem in different categories for network intrusion detection. In the first layer, we train a model for the multi-attack detection and find the class with the best accuracy. Next, the samples labelled as this class are removed while the rest of the data are used to train the model in the next layer. Similar to the previous layer, we train this model for the multi-attack detection and find the class with the best accuracy. This process is repeated until there are only three classes left for training. This model only focuses on the class that has the best performance in each layer. In the prediction phase, we apply the exact matching strategy layer by layer. Once the class is matched, we return the label as the final result; otherwise go to the next layer until a label is matched. The experimental results show that our proposed class-based strategy can improve the performance of the prediction models used. Among all the results, our class-based CatBoost has the best performance in terms of precision, recall, and F1-score where its accuracy is 0.76, the macro average of [precision, recall, F1-score] is [0.75, 0.71, 0.72], and the weighted average of [precision, recall, F1-score] is [0.75, 0.76, 0.75]. These results demonstrated that our class-based technique is effective for multi-attack prediction.