Volume 14 | Issue 6 |

 Abstract

Melancholy is a common mental disorder that has a significant impact on people's wellbeing and frequently results in serious emotional, cognitive, and physical deficits. More precise and scalable solutions are required since traditionaldiagnostic techniques, which rely on self-reported questionnaires and clinician interviews, are subject to subjectivity and prejudice. This work suggests a sophisticated machine learning-based method for using facial expression analysis to identify and categorize depression. By automatically evaluating and classifying depression severity, the process seeks to enhance the early detection of depressive disorders by utilizing cutting-edge deep learning algorithms. The study combines a number ofmachine learning models, such as EfficientNet, Vision Transformers, Visual Geometry Group (VGG16), and Residual Network (ResNet50). The goal of the suggested method is to automatically evaluate and categorize depression severity with more precision and dependability.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Depression detection, facial expression analysis, deep learning, emotion recognition.

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

This paper proposes a robust framework for detecting and classifying road surface defects--specifically cracks and potholes - using machine learning algorithms trained on annotated image datasets. High-resolution images of various road conditions are processed and fed into a CNN model, which learns visual features to differentiate between defect types and severities. Traditional inspection methods are labor-intensive, time-consuming, and subject to human error. With the emergence of computer vision and deep learning, particularly convolutional neural networks (CNNs), automated road surface analysis has become a practical solution. The rapid growth of urban infrastructure has made the maintenance of road surfaces a critical issue for city planners and civil engineers. Road cracks and potholes significantly contribute to traffic accidents and long-term infrastructure degradation. The system integrates pre-processing steps like image enhancement, edge detection, and data augmentation to improve detection accuracy under varied lighting and environmental conditions. The trained model achieves high precision in identifying surface anomalies, outperforming conventional techniques. Evaluation metrics such as accuracy, recall, and F1-score are used to validate performance. The proposed method offers scalable deployment options in real-time road surveillance systems through drones or vehicle-mounted cameras. Furthermore, the model supports predictive maintenance planning by pinpointing early-stage defects. This initiative reduces human effort, increases monitoring efficiency, and ultimately enhances road safety. The system's adaptability across diverse geographical terrains further highlights its practicality. With the integration of GPS and cloud storage, defect locations can be mapped and archived for future assessments. This AI-driven approach has the potential to revolutionize road maintenance and traffic safety management globally.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Computer Vision, Pavement Crack Detection, Pothole Identification and Segmentation, Deep Learning, Convolutional Neural Network (CNN), Image Classification, Surface Defect Detection, Edge Detection, Automated Inspection, Road Maintenance, , Predictive Maintenance, Real-Time Detection.

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

This paper presents a comprehensive review of deep learning-based approaches for real-time recognition of continuous sign language. The primary objective is to analyze and compare various techniques used in gesture recognition systems that translate sign language into text and speech, thereby enabling effective communication for deaf and hard-of-hearing individuals. The study focuses on computer vision-based models such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM), and Media Pipe-based hand tracking systems.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Sign Language Recognition, Deep Learning, CNN, LSTM, Media Pipe, Computer Vision, Real-Time Systems, Assistive Technology

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Pneumonia is a serious respiratory disease that affects millions of people worldwide and can be life-threatening if not diagnosed and treated early. Traditional diagnosis of pneumonia is primarily based on clinical examination and interpretation of chest X-ray images by radiologists. However, this manual process can be time-consuming, subjective, and prone to human error, especially in regions with limited access to expert healthcare professionals. To address these challenges, the application of Deep Learning, particularly Convolutional Neural Networks (CNNs), has emerged as an efficient and reliable approach for automated pneumonia detection. This project focuses on developing a computer-aided diagnostic system that utilizes CNN models to classify chest X-ray images as pneumonia-infected or normal. CNNs are a class of deep neural networks specifically designed for image processing tasks, capable of automatically extracting relevant features such as edges, textures, and patterns from medical images. In this study, a large dataset of labeled chest X-ray images is used to train the model, enabling it to learn distinguishing characteristics of pneumonia. The proposed system involves several stages, including image preprocessing, data augmentation, model training, validation, and testing. Preprocessing techniques such as resizing, normalization, and noise reduction are applied to improve image quality and enhance model performance. Data augmentation methods like rotation, flipping, and zooming are used to increase dataset diversity and prevent overfitting. The CNN architecture typically consists of multiple convolutional layers, pooling layers, and fully connected layers that work together to extract features and perform classification. The trained model is evaluated using performance metrics such as accuracy, precision, recall, and F1-score to ensure reliability and effectiveness. Experimental results demonstrate that CNN-based models can achieve high accuracy in detecting pneumonia from chest X-ray images, often outperforming traditional machine learning methods. This automated system can assist radiologists in making faster and more accurate diagnoses, thereby improving patient outcomes..

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Pneumonia Detection, Convolutional Neural Network (CNN), Chest X-ray Imaging, Deep Learning, Medical Image Processing, Image Classification etc.

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

This paper presents a multimodal AI-based arthritis detection system integrating five independent diagnostic modules: a Vision Module (knee X-ray classification, spine MRI degeneration analysis, and synovial fluid microscopy), a Lab Report Module, a Wearable Sensor Module, a Genomics Module, and a Voice Emotion Detection Module. The Vision Module employs DenseNet121 trained on 5,778 knee X-ray images, achieving 82.40% classification accuracy across five Kellgren-Lawrence (KL) grades. Spine MRI degeneration is assessed via signal intensity analysis, while synovial fluid inflammation is detected using ResNet50 with K-Means clustering. The Lab Report Module applies Random Forest classification on clinical biomarkers. The Wearable Module employs a neural network on MotionSense sensor data for risk stratification. The Genomics Module classifies arthritis-associated SNP markers, and the Voice Module detects pain and fatigue using MFCC-based feature extraction. All modules produce structured outputs integrated into an automated PDF medical report generator, providing a scalable, interpretable, and accessible clinical decision support system.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Arthritis detection, deep learning, DenseNet121, multimodal AI, Kellgren-Lawrence grading, ResNet50, MFCC, genomics, wearable sensors, clinical decision support

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Predicting the stock market has always been tricky, given how random, complicated, and unstable prices can be. Classification-based methods--those that forecast whether prices will go up or down--have become popular, mostly because they fit real trading strategies well. This paper takes a close look at classification techniques used in stock market prediction, from old-school machine learning models to ensemble methods and deep learning. We dig into how these approaches work, what they do well, where they fall short, and how suitable they are for analyzing financial time-series data. On top of that, we highlight major research gaps when it comes to temporal stability, feature interaction, and robustness, and point out promising directions for future studies.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Stock Market Prediction, Classification Models, Machine Learning, Deep Learning, Financial Time-Series.

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

The proposed system is based on e-certificate system in India's educational framework that leverages blockchain technology to address the widespread issue of certificate forgery. The inherent qualities of blockchain, including its immutability and transparency, create a solid foundation for enhancing the security and reliability of educational certifications. The operational framework is built around creating and storing an electronic file containing essential academic information in a dedicated database. Concurrently, the system generates a unique hash value for this electronic file, which serves as an exclusive identifier. This hash is securely embedded within a blockchain block, benefiting from the technology's resistance to tampering. To enable verification processes, both an inquiry string code and a QR code are linked to each certificate, encapsulating necessary information for authenticity verification. Users can initiate validation by scanning the QR code with a mobile device or entering the inquiry string on a specific website. The hash stored in the blockchain is then checked to confirm that the certificate has not been altered. This proposed solution significantly enhances the credibility of traditional paper certificates by implementing a dependable, transparent, and tamper-resistant verification process.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Blockchain, Digital Certificate, Hashing, E-Certificate, Certificate Verification, Transparency, QR Code, Secure Storage, Credential Authentication

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Ransomware has emerged as one of the most dam- aging cyber threats, targeting personal, enterprise, and critical infrastructure systems by encrypting data and demanding ran- som payments. Signature-based antivirus solutions fail to detect modern ransomware variants due to polymorphism, encryption, and obfuscation techniques. To address these limitations, this paper proposes a hybrid stacked machine learning framework for ransomware detection and classification. The proposed system integrates static and dynamic analysis to extract discriminative features such as Portable Executable headers, entropy values, API calls, file system operations, registry modifications, and network behavior. Feature selection is performed using an Extra Tree Classifier to reduce dimensionality and improve learning efficiency. A stacked ensemble architecture combining Extra Tree Classifier as the base learner and Logistic Regression as the meta learner is employed to enhance classification performance while maintaining interpretability and low computational overhead. The system is deployed using a Flask-based web interface for real-time file analysis. Experimental results demonstrate that the proposed approach achieves 98.2% accuracy with 1.2% false positive rate, outperforming traditional machine learning models and providing competitive performance compared to deep learning approaches with significantly lower computational cost, making it suitable for practical ransomware detection.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Ransomware Detection, Malware Classification, Hybrid Machine Learning, Stacked Ensemble, Extra Tree Classi- fier, Logistic Regression, Cybersecurity, Static Analysis, Dynamic Analysis

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Efficient garbage separation is greatly essential in modern-day waste management systems and has direct implications for the efficiency of recycling, environmental sustainability, and public health. The classical process of manual garbage separation is cumbersome and prone to errors, as well as exposed to harmful materials. With the ever-increasing rate of urbanization and the consequent rise in consumer waste, there is an immense need for intelligent and automated solutions to effectively classify the garbage in a precise and minimally supervised manner. This paper describes Automated Garbage Classification based on Machine Learning (ML) and Deep Learning (DL) techniques to classify images of garbage into pre-defined classes of plastic, paper, metal, battery, and general trash. The project employs multiple techniques of classification like K-Nearest Neighbours (KNN), Support Vector Machine (SVM), and Random Forest Classification, along with a Convolutional Neural Network (CNN) to analyze and compare performance of each technique. Additionally, image processing techniques of resizing, normalizing images, and data augmentation are performed. Results reveal that the CNN approach outperforms classical machine learning techniques in terms of accuracy and reliability, especially in adverse conditions of light and background noise. The system is web-developed and based on Python Flask, enabling users to classify images for real-time results and confidence scores.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Waste Segregation, Garbage Classification, Convolutional Neural Network (CNN), Machine Learning, Sustainability, Deep Learning, Image Classification.

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

CE-39 Feature-Model Synergy in Cry-Based Detection of Neonatal Asphyxia Using Hybrid Cepstral and Neural Approaches

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

CE-39 Feature-Model Synergy in Cry-Based Detection of Neonatal Asphyxia Using Hybrid Cepstral and Neural Approaches

  License

Creative Commons Attribution 4.0 and The Open Definition