Abstract The rapid development of Large Language Models (LLMs) has made it increasingly hard to distinguish human written from AI-generated academic papers, raising questions about authorship, originality, and academic integrity. This research presents a systematic approach to identifying... The rapid development of Large Language Models (LLMs) has made it increasingly hard to distinguish human written from AI-generated academic papers, raising questions about authorship, originality, and academic integrity. This research presents a systematic approach to identifying and measuring multidimensional markers that differentiate human written academic manuscripts from AI-generated and AI-paraphrased summaries. After analysing 52 human written academic papers and their AI-generated versions (created with GPT-4o, DeepSeek V4, and Qwen 3.7plus), we developed an automated, multi-layered text-annotation tool. It incorporates rule based matching, zero shot classification, and structured LLMaided annotation, and we validated it using human inter annotator agreement. We found that no single language marker is enough for classification. However, the accurate classification (85-90%) requires a set of markers which include the completeness of the scholarly apparatus (the most reliable one), significant information loss (40-50% of granular details such as exact statistics or software name), decrease of syntactic burstiness (by 40-50%), formulaic transition substitutions, and the drop in epistemic hedging by 60%. Although the absence of structure and metadata yields perfect classification, the stylometric approach alone produced an AUC of 0.78, underscoring the complexity of linguistic signatures. In summary, successfully implementing AI for identifying text in academic environments requires a multi marker approach that favours completeness, informativeness, and text depth over fluency and vocabulary sophistication, which are highly prone to false positive results. Previous research has addressed AI academic detection; however, we shift our emphasis from surface fluency (which LLMs excel at mimicking) to epistemic rigour and granular information loss. Read More Read Less

Abstract Social media contains opinions, ideas, and facts. Artificial intelligence (AI) has brought a mix of societal perspectives to social media events. This study used bilingual tweets (English and Malay). This study uses Malay stop words and number... Social media contains opinions, ideas, and facts. Artificial intelligence (AI) has brought a mix of societal perspectives to social media events. This study used bilingual tweets (English and Malay). This study uses Malay stop words and number filters in data preprocessing, builds a machine learning classifier using mBERT, support vector machine, and neural network models, and measures performance using recall, precision, Fmeasure, and accuracy for each model. In the experimental results, Malay stop words and number filters caused variability in classifier performance. Malay stop words and number filters are important in preprocessing student sentiment analysis in this study, although they do not significantly affect the change in accuracy. This paper also discusses the KNIME workflow and the experimental results obtained. This study argues that the neural network is the best classifier, and the model's algorithm can be used to perform sentiment analysis on a new dataset. Read More Read Less

Abstract This paper studies how noise and decoherence affect a hybrid quantum classical neural network (HQCNN) when it is applied to EEG based eye state classification, a task that closely mirrors what an edge based braincomputer interface would... This paper studies how noise and decoherence affect a hybrid quantum classical neural network (HQCNN) when it is applied to EEG based eye state classification, a task that closely mirrors what an edge based braincomputer interface would need to do in the field. Quantum machine learning holds real promise for embedded health systems, but the hardware of today is noisy, and nobody has clearly mapped out exactly where the accuracy starts to collapse. We built a model that uses a four qubit parameterized quantum circuit implemented in PennyLane and backed by a PyTorch classical front end, then trained it on the wellknown EEG Eye State dataset from OpenML. Once training was done, we deliberately injected depolarizing channel noise at six different error rates from a clean 0.00 all the way up to 0.15 and measured what happened to classification accuracy at each step. The results gave us a clear picture. The noiseless model reached 90.75% accuracy. That number held reasonably steady through low noise, but at a depolarizing error rate of around 0.12 the accuracy dropped below the 90% threshold of the baseline, landing at 86.00%, and by 0.15 it had fallen further to 83.75%. We define a hard mathematical threshold as a 10% relative degradation from the noiseless baseline (0.8168). However, the model did not reach this absolute floor at the maximum tested noise of 0.15 (83.75%). Instead, we identify 0.12 as the practical decoherence warning threshold, defined by a sharp inflection in the rate of change (slope) of the accuracy drop. This distinction clarifies that while the model remains mathematically above the 10% degradation floor, 0.12 represents the critical point where hardware engineers must intervene before clinical deployment. The work proves that a compact four qubit circuit can remain useful for EEG classification even under moderate hardware noise, while also making it clear that current NISQ devices would need error mitigation strategies before being trusted in a clinical edge deployment. The findings contribute a concrete benchmark that system designers can use when deciding whether quantum co-processors belong in the next generation of wearable EEG telemetry hardware. Read More Read Less