The facial features of children can provide important visual cues for the early detection of autism spectrum disorder (ASD). This research focuses on developing an image-based detection system to identify children with ASD. The main problem addressed is the lack of practical methods to assist healthcare professionals in the early identification of ASD through facial visual characteristics. This study aims to design a prototype facial image acquisition and detection system for children with ASD using Raspberry Pi and a deep learning-based single shot detector (SSD) algorithm. In this method, the face detection model uses a modified ResNet50 architecture, which can be used for advanced analysis for classification between autistic and normal children, achieving 95% recognition accuracy on a dataset consisting of facial images of children with and without ASD. The system is able to recognize the visual characteristics of the faces of children with ASD and consistently distinguish them from those of normal children. Real-time testing shows a detection accuracy ranging from 86% to 90%, with an average accuracy of 90%, despite fluctuations caused by variations in movement and viewing angle. These results show that the developed system offers high accuracy and has the potential to function as a reliable diagnostic tool for the early detection of ASD, which ultimately facilitates timely intervention by healthcare professionals to support the optimal development of children with ASD.

The classification of nutritional status in children under two years old is crucial for monitoring growth and early detection of nutritional problems. However, in many healthcare facilities, this classification is still performed manually, requiring relatively long processing times and being prone to human error in both measurement and data recording. The problem addressed in this study is the inefficiency and potential inaccuracy of manual nutritional status classification in toddlers. This research aims to develop an automatic and digital device capable of measuring body length and weight and classifying nutritional status in children under two years old efficiently, accurately, and in real time. The device utilizes electronic sensors integrated with a microcontroller to streamline the process and reduce measurement error. The main contribution of this study is the design and realization of a portable automation device that integrates an HC-SR04 ultrasonic sensor for measuring body length and a 50 kg full-bridge load cell sensor for measuring body weight, both controlled by an ATmega328P microcontroller. The device processes the data measurement digitally, displays the results on a 20 × 4 LCD, and provides a printed copy via a thermal printer, enhancing the data recording efficiency. The method involves the design of hardware circuits, sensor calibration, software programming using the C language in the Arduino IDE, and performance testing of the device by comparing its results to standard measuring instruments. The device’s performance is evaluated based on measurement error percentage and precision level. The results demonstrate that the device achieved an error percentage of 1.26% for body length measurement and 0.98% for body weight measurement. The overall system error is recorded at 0.5%, with a precision level ranging from ±0.08 to ±0.4.

Manual annotation of large datasets is a time consuming and resource-intensive process. Hiring annotators or outsourcing to specialized platforms can be costly, particularly for datasets requiring domain-specific expertise. Additionally, human annotation may introduce inconsistencies, especially when dealing with complex or ambiguous data, as interpretations can vary among annotators. Large Language Models (LLMs) offer a promising alternative by automating data annotation, potentially improving scalability and consistency. This study evaluates the performance of Chat GPT compared to human annotators in annotating an Islamophobia dataset. The dataset consists of fifty tweets from the X platform using the keywords Islam, Muslim, hijab, stop islam, jihadist, extremist, and terrorism. Human annotators, including experts in Islamic studies, linguistics, and clinical psychology, serve as a benchmark for accuracy. Cohen’s Kappa was used to measure agreement between LLM and human annotators. The results show substantial agreement between LLM and language experts (0.653) and clinical psychologists (0.638), while agreement with Islamic studies experts was fair (0.353). Overall, LLM demonstrated a substantial agreement (0.632) with all human annotators. Chat GPT achieved an overall accuracy of 82%, a recall of 69.5%, an F1-score of 77.2%, and a precision of 88%, indicating strong effectiveness in identifying Islamophobia related content. The findings suggest that LLMs can effectively detect Islamophobic content and serve as valuable tools for preliminary screenings or as complementary aids to human annotation. Through this analysis, the study seeks to understand the strengths and limitations of LLMs in handling nuanced and culturally sensitive data, contributing to broader discussion on the integration of generative AI in annotation tasks. While LLMs show great potential in sentiment analysis, challenges remain in interpreting context-specific nuances. This study underscores the role of generative AI in enhancing human annotation efforts while highlighting the need for continuous improvements to optimize performance.

Islamophobia remains a globally pervasive issue, but its local manifestations often reflect unique socio-cultural and political contexts. In Muslim-majority countries like Malaysia, Islamophobia is less about minority marginalization and more about internalized stereotyping, religious politicization, and cross-cultural misunderstanding. Despite growing discourse, localized and experiential analyses of Islamophobia in Southeast Asia remain limited. This study aims to explore how Malaysian Muslims perceive and experience Islamophobia through qualitative responses. A total of 154 open-ended survey responses were collected and analysed using thematic analysis and word cloud visualization to enhance the interpretation of prominent keywords. The study revealed four dominant themes: Misunderstanding of Islam, Negative Stereotyping, Institutional or Politically Driven, and Social Rejection and Discrimination. These themes illustrate the multifaceted nature of Islamophobia, ranging from conceptual misperceptions of Islam and gendered stereotypes to anxieties surrounding Islamic governance and personal experiences of exclusion, particularly among converts. The findings contribute to Islamophobia literature by focusing on Muslim opinions within a non-Western, Muslim-majority setting, thereby challenging the predominantly Western-centric framing of the phenomenon. The findings of this study highlight the influence of media, identity politics, and family dynamics in shaping Islamophobic attitudes in Malaysia. Implications include the need for culturally responsive policies, inclusive religious education, and public discourse reform to counteract Islamophobic narratives. In conclusion, the research underscores the importance of context-sensitive approaches in Islamophobia studies and calls for further comparative investigations across diverse geopolitical and religious environments

In recent years, significant advancements in facial recognition technology have been marked by the prominent use of convolutional neural networks (CNN), particularly in identification applications. This study introduces a novel approach to face recognition by employing ResNet-50 in conjunction with the DeepLabV3 segmentation method. The primary focus of this research lies in the thorough analysis of ResNet-50's performance both without and with the integration of DeepLabV3+ segmentation, specifically in the context of datasets comprising faces of children on the autism spectrum (ASD). The utilization of DeepLabV3+ serves a dual purpose: firstly, to mitigate noise within the datasets, and secondly, to eliminate unnecessary features, ultimately enhancing overall accuracy. Initial results obtained from datasets without segmentation demonstrate a commendable accuracy of 83.7%. However, the integration of DeepLabV3+ yields a substantial improvement, with accuracy soaring to 85.9%. The success of DeepLabV3+ in effectively segmenting and reducing noise within the dataset underscores its pivotal role in refining facial recognition accuracy. In essence, this study underscores the pivotal role of DeepLabV3+ in the realm of facial recognition, showcasing its efficacy in reducing noise and eliminating extraneous features from datasets. The tangible outcome of increased accuracy of 85.9% post-segmentation lends credence to the assertion that DeepLabV3+ significantly contributes to refining the precision of facial recognition systems, particularly when dealing with datasets featuring faces of children on the autism spectrum. Downloads

Traffic accidents often result in loss of life and significant economic losses. Indonesia's high number of traffic accidents indicates the need for effective solutions to overcome this problem. Developing a drowsiness detection device is one effort that can be made to reduce accidents caused by drowsy drivers. The data obtained in this study used driver heart rate data. The drowsiness detection tool was developed using the Wemos D1 Pro Esp8266 microcontroller and MAX30102 sensor. Testing was carried out on 25 subjects under two conditions: 'Drowsy' and 'Normal.' The driver's level of drowsiness is determined based on the heart rate measured by the detection device. The Blynk application is used as a visual interface to provide notifications via smartphone if the driver is drowsy. The accuracy of the drowsiness detection tool was compared with the results obtained from the Pulse Oximeter. This research shows that the drowsiness detection tool using the Wemos D1 Pro Esp8266 microcontroller and MAX30102 sensor has an accuracy of around 98% when compared with the pulse oximeter. The Blynk application successfully sends notifications precisely when the driver is drowsy. This study highlights the potential of drowsiness detection devices to improve traffic safety and reduce accidents caused by drowsy drivers.

Islamophobia is commonly understood as prejudice, irrational fear, discrimination, or hostility directed toward Islam and Muslims. Although global scholarship on Islamophobia has grown, the discourse remains predominantly Western-centric and often lacks the cultural nuance needed to capture how Islamophobia manifests in Muslim-majority settings such as Malaysia. This study addresses this gap by examining Islamophobia discourse in Malaysia using a survey-based keyword analysis anchored in local sociocultural perspectives. Data were gathered from 229 participants in Melaka through a structured survey designed to represent a broad demographic range, including variations in age, gender, education level, and ethnicity. An initial set of 39 Islamophobia-related keywords was derived from prior academic literature, media narratives, and Malaysian socio-political discussions. These keywords were subsequently reviewed and validated by three subject-matter experts in religious studies, media and communication, and semantic web and ontology to ensure contextual relevance and semantic precision. Quantitative analyses were conducted to examine the frequency, clustering, and categorization of the validated keywords using the Tripartite Islamophobia Scale, which conceptualizes Islamophobia along three dimensions: Anti-Muslim Prejudice, Anti-Islamic Sentiment, and Conspiracy Beliefs. The findings show that most keywords align with the Anti-Islamic Sentiment dimension. Frequently recurring terms such as “Halal,” “Allah,” “Syariah law,” “Penyalahgunaan Kalimah Allah,” and “attacking Islam” emerged as central to negative or politicized portrayals of Islam and Muslims in the Malaysian context. The results further indicate notable demographic differences in perceptions, suggesting that communities interpret and experience Islamophobia in diverse ways. A key contribution of this study is the development of a context-sensitive keyword set tailored specifically to Malaysia. This set provides a foundational tool for systematically extracting and analysing textual data for constructing a comprehensive Islamophobia corpus. By grounding keyword selection in empirical validation and local sociocultural insights, the study advances corpus-building methodologies and offers a valuable resource for future research, policy development, media monitoring, and public education initiatives across Southeast Asia.

Electroencephalography (EEG) is a non-invasive technique for monitoring and recording the brain's electrical activity with electrodes applied to the scalp. The method is important in neurological studies, like that of Autism Spectrum Disorder (ASD), because it measures patterns of brain waves that can identify developmental abnormalities. However, EEG signals are often contaminated by multiple noise sources, including eye movements, muscle activity, and extraneous interference. This interference can significantly reduce the quality and intelligibility of signals. Therefore, preprocessing is required to enhance the reliability and precision of the data obtained. In this study, a Butterworth Band-Pass Filter (BPF) was used during preprocessing to filter out undesirable frequency components and to mitigate noise. After filtering, EEG signals were handled using the Variational Mode Decomposition (VMD) technique. VMD is an adaptive method for decomposing multidimensional signals into intrinsic mode functions while preserving critical details of the original data. For performance comparison, four quantitative metrics were used: Mean Squared Error (MSE), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Signal-to-Noise Ratio (SNR). Results showed that VMD performed better than BPF alone. As an example, for Subject 1, VMD achieved an MAE of 0.26 and MSE of 0.42, which was far superior to the MAE of 13.72 and MSE of 674.96 of BPF. Subject 3 had the least RMSE (0.40) when using VMD, whereas BPF scored 25.90. VMD also reported a highest SNR of 28.56, compared to BPF's 2.43. Overall, integrating VMD with BPF significantly improves EEG signal quality and enables more accurate analysis, particularly in ASD-related studies.

This study discusses the face recognition of children with special needs, especially those with autism. Autism Spectrum Disorder (ASD) is a complex neurodevelopmental disorder that affects social skills, ways of interacting, and communication disorders. Facial recognition in autistic children is needed to help detect autism quickly to minimize the risk of further complications. There is extraordinarily little research on facial recognition of autistic children, and the resulting system is not fully accurate. This research proposes using the Convolution Neural Network (CNN) model using two architectures: ShuffleNet, which uses randomization channels, and Visual Geometry Group (VGG)-19, which has 19 layers for the classification process. The research object used in the face recognition system is secondary data obtained through the Kaggle site with a total of 2,940 image data consisting of images of autism and non-autism. The faces of autistic children are visually difficult to distinguish from those of normal children. Therefore, this system was built to recognize the faces of people with autism. The method used in this research is applying the CNN model to autism face recognition through images by comparing two architectures to see their best performance. Autism and non-autism data are grouped into training data, 2,540, and test data, as much as 300. In the training stage, the data was validated using validation data consisting of 50 autism image data and 50 non-autism image data. The experimental results show that the VGG-19 has high accuracy at 98%, while ShuffleNet is 88%.

Stress is prone to occur in children with autism. According to the study, around 85% of children who have autism suffer from anxiety disorders that can exacerbate their condition, leading to self-harm and harm to those in their vicinity. Heart rate, skin conductance, and finger temperature changes occur during stress. In this paper, we design a system to monitor heart rate, body temperature, and skin conductance to detect signs of stress. Subsequently, the measurement data is processed using the fuzzy logic (FL) method as a decision-maker algorithm. In particular, we use 64 fuzzy rules with membership functions for each parameter. Parameter measurement results will be displayed using a widget called Gauge, while stress conditions will be displayed using a label widget. The results will be displayed on the Blynk application with an IoT system and viewed remotely via Android devices. The test was conducted on five children aged 5-9 years with varying body conditions. From the test results, the mean accuracy of the heart rate sensor was 95.01%, the mean temperature sensor accuracy was 97.7%, and the mean conductance sensor accuracy was 93.75%. The stress levels range from a minimum of 25% to a maximum of 75%. These findings indicate that the developed tool has performed effectively, and it is feasible to monitor its operation remotely.

In autistic children, one of the important physiological aspects to be examined is the heart condition, which can be assessed through electrocardiogram (ECG) signal analysis. However, ECG signals in autistic children often contain interference in the form of noise, making the analysis process, both manual and conventional, challenging. Therefore, this study aims to analyze the ECG signals of autistic children using a classification method to distinguish between two main conditions: playing and calm conditions. A deep learning approach employing the Convolutional Neural Network (CNN) architectures was used to obtain accurate results in distinguishing the heart conditions of autistic children. The data used consists of 700 ECG signal data in each class, processed through the filtering, windowing, and augmentation stages to obtain balanced data.  Three CNN architectures, ResNet, DenseNet, and XceptionNet, were tested in this study. Although these architectures are originally designed for 2D and 3D image data, modifications were made to adapt the input data structure to perform 1D data calculations. The evaluation results show that the XceptionNet model achieved the best performance, with accuracy, precision, recall, and F1-score of 97,14% each, indicating a good ability in capturing the complex patterns of ECG signals. Meanwhile, the ResNet obtained good results with 96,19% accuracy, while DenseNet performed slightly lower results with 94,76% accuracy and evaluation metrics. Overall, this study demonstrates that a deep CNN architecture based on dense connections can enhance the accuracy of ECG signal classification in autistic children.

Electroencephalography (EEG) measures electrical activity in the brain area by placing several electrodes on the scalp that can be used to diagnose autism spectrum disorder (ASD) and various abnormalities in the brain nerves. During the EEG signal recording process, the measured signal is often contaminated by various types of noise, which causes difficulties in analyzing the signal. Therefore, an effective method is needed to reduce these artifacts. This research applied wiener filter (WF) and savitzky-golay filter (SG) methods in reducing noise in the EEG signals of autistic people. This method will be combined with another method, namely Butterworth Band-Pass Filter, to concentrate the frequency in the range of 0.5-40 Hz. Based on the comparison of performance accuracy values using three calculation parameters, namely mean square errors (MSE), Mean absolute errors (MAE), and signal to noise ratio (SNR), this study proves that WF is superior to SG in producing EEG signals of autistic and normal people free from noise. WF shows an SNR value of 34.773 "dB" compared to 22.157 "dB" in SG, as well as lower MAE and MSE values of 0.521 μV and 0.616 μV2 compared to 1.875 μV and 16.990 μV2 in SG. These results confirm that WF is more effective in reducing noise interference and producing more accurate signal estimation in EEG data analysis.

The increasing incidence of minors accessing hazardous indoor areas—such as staircases, balconies, and rooms with sharp objects—raises serious safety concerns, often due to insufficient parental supervision. This study proposes an Internet of Things (IoT)-based automatic door lock system to enhance child safety in home environments. The system integrates dual ultrasonic sensors for distance and height detection, a KY-037 sound sensor, and an ESP32-CAM for real-time video monitoring, all accessible via a web interface. A key novelty lies in the integration of multi-sensor spatial awareness with live surveillance, enabling automated control and proactive safety features. Tested on ten children aged 4 to 6 years, the system achieved a 90% success rate in locking the door when a child under 120 cm approached within 1 meter, with an average response time of approximately 2 seconds. A sound-based alarm is also triggered when noise levels exceed 120 decibels, serving as an emergency alert. However, a 10% false negative rate was observed when children were detected at distances of 1.3 to 1.5 meters, suggesting the need for further sensor calibration. Overall, the system demonstrates strong potential as a scalable and cost-effective smart home safety solution, combining automation, real-time monitoring, and child-specific access control. Future work should focus on improving detection accuracy and extending functionality for multi-object scenarios.

Autism spectrum disorder is a serious mental disorder affecting social behavior. Some children also faceintellectual delay. In people with autism spectrum disorder, the signals detected have abnormalities compared to normalpeople. This can be a reference in diagnosing the disorder with electroencephalography. This study will analyze the effect ofpower spectral density on the electroencephalography of autistic children and also compare it with the power spectral densityvalue on the electroencephalography of normal children using the Welch periodogram method approach. In the preprocessingstage, the independent component analysis method will be applied to remove artifacts, and a finite impulse response filter toreduce noise in the electroencephalography signal. The study results indicate differences in the power spectral density valuesobtained in the autistic and normal electroencephalography signals. The power spectral density value obtained in the autisticelectroencephalography signal is higher than the normal electroencephalography signal in all frequency sub-bands. From thestudy results, the highest power spectral density value obtained by the autistic electroencephalography signal is in the deltasub-band, which is 54.06 dB/Hz, while the normal electroencephalography signal is only 33.14 dB/Hz at the same frequencysub-band. And in the Alpha and Beta sub-bands, the normal electroencephalography signal increases the power spectraldensity value, while in the autistic electroencephalography signal, the power spectral density value decreases in the Alphaand Beta sub-bands. In addition, finite impulse response and independent component analysis methods can also reduce noiseand artifacts contained in autistic and normal electroencephalography signals

During pandemic COVID-19 outbreaks, the number of ransomware cyber-attacks has increased tremendously. Ransomware often spreads through malicious links or compromised URLS. Such link infects a computer and restricts users’ access until a ransom is paid to unlock it. Nowadays, several ransomware detection techniques have been developed, however these approaches were either unsuccessful or unable to prevent these attacks. One of the downsides is due to the poor detection accuracy and low adaptability to the new variants of ransomware. Another reason behind the unsuccessful ransomware detection solutions is an arbitrary selected URL-based classification features which may produce false results to the detection. The objective of this research is to identify the potential URL-based web characteristics that can be manipulated to infect computers. Therefore, in this paper, twelve (12) potential URL-based web characteristics that can be manipulated to infect computers were analysed. Datasets of 10000 legitimate emails and a number of 3240 phishing emails were examined using heuristic-based model. It is believed that, an ongoing updated ransomware rules and database which is built-on immunological memory of updated URL ransomware features will prevent ransomware attacks with an appropriate countermeasure for prevention and removal.