VOLUME 15, ISSUE 4, APRIL 2026

Abstract: Child abuse remains a major global concern that threatens the physical, emotional, and psychological development of children. Despite growing awareness and legal frameworks aimed at protecting children, many abuse cases remain undetected or are reported too late due to inadequate monitoring systems, poor data quality, fear of stigma, and ineffective reporting mechanisms. Psychological abuse is often difficult to detect because it does not leave visible physical evidence and is therefore frequently underreported. In addition, the rapid expansion of digital communication platforms has created new avenues through which abusive behaviors such as cyberbullying, harassment, and threats can occur, making traditional monitoring approaches insufficient. This study presents the development of the Child Abuse Monitoring and Reporting (CAMR) System that uses Natural Language Processing (NLP) to enhance the detection, monitoring, and reporting of child abuse cases. The proposed system employs sentiment analysis and Named Entity Recognition (NER) techniques to identify emotional tones and relevant entities in text that may indicate abusive interactions. The system is implemented using the Naïve Bayes algorithm to classify text and detect potentially abusive content.

The system's performance is evaluated using accuracy, precision, recall, and F1-score, along with User Acceptance Testing (UAT), to assess its effectiveness and usability. The study demonstrates that integrating NLP techniques into child protection systems can enhance early detection of abuse, enable automated monitoring of large volumes of textual data, and support timely intervention. The proposed system contributes to improving child protection strategies by providing a scalable and efficient technological solution for monitoring and reporting abuse, particularly within digital communication environments. Ultimately, the system supports government agencies, institutions, and child protection organizations in safeguarding children and responding more effectively to abuse cases.

Keywords: Child Abuse; Child Protection; Abuse Detection; Psychological Abuse; Cyberbullying; Harassment; Threats.

Abstract: Modern computing systems are constantly exposed to unknown and evolving threats, making it difficult to ensure reliable protection using traditional security methods alone. While machine learning-based approaches have improved the ability to detect malicious activities, many of these systems still fail to clearly explain their decisions or respond quickly enough when a threat is identified. As a result, there is often a gap between detection, understanding, and action.

This paper presents a real-time malware detection framework that focuses on explainability, traceability, and automated response. The proposed system monitors system-level behavior and analyzes process activities using a transformer-based model that captures patterns over time. When a process is identified as suspicious, the system provides a clear, humanreadable explanation describing why it is considered malicious, along with traceable details such as where the activity originated and how it progressed within the system.

To minimize the impact of potential threats, the framework includes an automated response mechanism. If a process exceeds a defined risk threshold based on abnormal behavior, it is immediately terminated or isolated. In addition, a structured report is generated and stored, allowing users or analysts to review the complete details of the event whenever required.

Unlike existing approaches that treat detection and response separately, this work integrates detection, explanation, and action into a single unified framework. This not only reduces response time but also improves the clarity and usability of the system, making it more practical for real-world cybersecurity scenarios.

Furthermore, the system is designed to operate in real time without introducing significant overhead, ensuring that it remains efficient even in dynamic environments. By combining accurate detection with clear explanation and immediate response, the proposed approach aims to improve both trust and effectiveness in modern malware defense systems.

Keywords: Malware Detection, Explainable AI, Behavioral Analysis, Transformer Models, Real-Time Monitoring, Automated Response, Traceability, Cybersecurity

Abstract: For ecological preservation and environmental management, monitoring biodiversity in forest ecosystems is essential. Conventional monitoring techniques, like camera traps and direct observation, are frequently labor-intensive, time-consuming, and have a limited geographic and temporal reach. In this paper, we propose SoundForest, an AI-powered system that analyse natural ambient audio recordings to classify and track forest biodiversity. The system continuously gathers soundscape data from forest environments using edge or remote sensors. After preprocessing to eliminate background noise, the audio data is converted into time-frequency representations like Mel-Frequency Cepstral Coefficients (MFCCs) and Mel spectrograms. Deep learning models, namely Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Audio Spectrogram Transformers (AST) for multi-class sound classification, are trained using these representations. The models can detect changes in acoustic biodiversity over time and recognize vocalizations specific to a species. Anthropogenic sounds, such as chainsaw activity, gunshots, car engines, and human voices, are also recognized by SoundForest in order to detect wildlife and warn of unlawful logging, poaching, or unauthorized entry into protected forest areas. Even in remote locations, real-time, decentralized monitoring is made possible by the system’s support for deployment on low-power edge devices like Raspberry Pi. Supported by the system are heatmaps of biodiversity, behavioural patterns, and species distributions.

Keywords: Biodiversity Monitoring, Acoustic Sensing, Environmental Sound Classification, Deep Learning, CNN, RNN, Audio Spectrogram Transformer, Wildlife Conservation

Abstract: Customer churn represents a critical business challenge in telecommunications and subscription-based services, with annual revenue losses exceeding billions of dollars globally. This research presents a comprehensive machine learning framework for predicting customer churn using an ensemble of five advanced algorithms: Logistic Regression, Ran-dom Forest, XGBoost, LightGBM, and Gradient Boosting. We integrate structured behavioral data with sentiment analysis derived from customer feedback using TextBlob, incorporating dynamic sentiment weighting that adjusts churn probabilities by up to 45%. The framework employs SMOTE (Synthetic Minority Over-sampling Technique) for handling class imbalance and SHAP (SHapley Additive Explanations) for model interpretability. Experimental results on a telecommunications dataset of 7,043 customers demonstrate that the XGBoost classifier achieves supe-rior performance with 89.4% accuracy, 0.86 F1-score, and 0.95 AUC-ROC, outperforming baseline models by 7.8%. The sentiment weighting mechanism reduces false negatives by 23%, significantly improving identification of at-risk customers. The complete system is deployed as an interactive Gradio web application with real-time sentiment analysis, enabling businesses to make data-driven retention decisions. This research contributes a production-ready, interpretable churn prediction system that integrates multiple ensemble methods with sentiment-based probability adjustment for enhanced predictive accuracy.

Keywords: Customer Churn Prediction, Ensemble Learning, XGBoost, LightGBM, Random Forest, SMOTE, Senti- ment Analysis, SHAP, Gradio Deployment

Abstract: This paper introduces an AI-driven Interview Preparation System, which aims to improve the candidate preparation to improve recruitment workflows by intelligent automation and multimodal analysis. The system incorporates a progressive, multi-level interview engine that enables the injection of context-aware questions using cutting-edge language models to allow the candidate to experience various structured interview situations that range from the basic HR interaction to a technical-level expert. To better the quality of assessments, the platform uses video- based real-time analysis of emotions and audio for response analysis to offer end-to-end evaluation on multiple dimensions such as relevance, clarity, confidence, and technical accuracy.

The architecture designed relies on a modular architecture comprising user interface using web and high-performance backend framework to handle the scalable processing. The frontend offers country-specific role-based dashboards for candidates, HR professionals, and administrators with various supporting features, such as scheduling interviews, tracking performance, and managing users. The backend makes use of the AI services for question generator and answer evaluation services whereas, on the other hand, facial emotion analysis models allows for better behavioural insights during the interview sessions. Additionally, semantic matching methods are also used to match candidate profiles to job requirements, making it possible to make better hiring decisions.

Experimental observations show that the system simulates realistic interview situations and at the same time offers usable feedback in form of detailed reports of the performance. The merging of multimodal information sources adds great value in evaluation depth that is not found within the text-based systems. Furthermore, the platform guarantees flexibility by configurable parameters such as levels of the interviews, thresholds for scoring and user roles, making it adaptable to different recruitment cases. The system plays a role in filling the gap between traditional methods of interview preparation along with modern artificial intelligence solutions by providing an end-to-end intelligent interview ecosystem.

Keywords: Artifical intelligence, interview preparation system, multimodal analysis, emotion recognition, natural language processing, candidate evaluation, machine learning, human resource management, semantic matching, fastapi.

Abstract: Users in today’s digital marketplace often struggle to verify the authenticity of products due to the rapid growth of counterfeit goods and the lack of reliable verification systems. Most existing product authentication methods rely on centralized databases or manual verification processes, which are vulnerable to data manipulation and lack transparency. These systems do not effectively provide secure traceability, product verification, and consumer trust within a single platform. This paper proposes a blockchain-based fake product detection system that utilizes decentralized technology to improve product authentication and supply chain transparency. The system records product information such as manufacturing details, batch number, and ownership history on a blockchain network, ensuring that the data remains secure and immutable. Consumers can verify the authenticity of products by scanning a QR code or unique product identifier linked to the blockchain record. In addition, the system enables stakeholders to track product movement across different stages of the supply chain. This helps in preventing counterfeit distribution and allows reliable product verification. Overall, the design demonstrates how a transparent and secure blockchain-based system can support better product authentication, consumer trust, and protection against counterfeit goods.

Keywords: Blockchain technology, Fake product detection, Product authentication, Counterfeit prevention, Supply chain transparency, QR code verification, Decentralized ledger system, Product traceability, Consumer trust, Secure product verification.

Abstract: This project introduces Safe Drive IoT, a crucial vehicle security and safety system engineered to directly confront drunk driving, a leading cause of road fatalities. The core design integrates an IoT framework with essential hardware components: an alcohol detection sensor, a GPS module, a GSM communication module, and an ignition locking mechanism. The system's operation is straightforward and uncompromising: upon vehicle entry, the alcohol sensor monitors the driver's breath. If the detected alcohol concentration surpasses the safety limit, the system immediately triggers a warning via an integrated buzzer and, most importantly, activates the ignition lock, preventing the vehicle from starting.

In the event of an attempted start under intoxicated conditions, the system utilizes the GPS module to determine the vehicle's precise location and the GSM module to transmit an instant, geolocated alert (SMS) to emergency contacts and enforcement agencies. Safe Drive IoT provides a robust, real-time safety layer that effectively enforces sobriety behind the wheel, ensuring that a simple, yet life-saving, mechanism governs vehicle operation.

Keywords: Drunk driving, IoT, alcohol detection, GSM, MQ-3, GPS.

Abstract: The increasing generation of waste demands efficient and accurate segregation methods. This project presents an IoT-based Automated Waste Segregation System that classifies waste into four categories: wet, dry, plastic, and metal. The system uses an ESP32 microcontroller integrated with moisture, inductive proximity, infrared, and ultrasonic sensors for detection and monitoring.

A conveyor belt mechanism with stepper and servo motors enables automatic sorting based on multi-stage sensing. The system also uses MQTT and HiveQL cloud for real-time monitoring through a web dashboard. Key advantages include low cost, reduced human effort, improved accuracy, and efficient handling of biodegradable wet waste.

This solution supports smart waste management and contributes to better recycling and environmental sustainability.

Abstract: This paper presents MediCard+, an AI-driven centralized healthcare system designed to address fragmented medical records and inefficient clinical decision-making. The platform enables secure storage, retrieval, and management of patient records across multiple hospitals using role-based access control. A key feature is an AI- powered chatbot that uses Natural Language Processing (NLP) and a pre-trained LLaMA-based model to analyze medical data and provide real-time insights. It supports medical history summarization, detection of critical conditions, and prediction of long-term disease risks. To ensure privacy, MediCard+ includes patient-controlled data sharing and a privacy-preserving AI pipeline that prevents direct exposure of sensitive data. The system uses MongoDB for structured data, cloud storage for documents, and a Node.js backend for scalability. Experimental results show improved data accessibility, faster response time, and better clinical decision support, making MediCard+ a scalable and secure solution for modern healthcare systems.
.

Keywords: Centralized EHR, Clinical Decision Support, Healthcare AI Chatbot, Role-Based Access Control, Medical Data Security, NLP in Healthcare

Abstract: - Facial Emotion Recognition, or FER, is at the heart of affective computing these days. It isn’t just making tech feel more human—it’s changing how we watch for mental health swings and beef up security, too. There’s a lot of excitement about what FER can do, but using these systems outside the lab brings some real headaches. Deep learning eats up a lot of computing power, and life is messy—dim rooms, people looking away, turning their heads. All that can trip up even solid models.

That’s what this study digs into. We put together a real-time emotion detection system using DeepFace and OpenCV. Everything hinges on deep CNNs—they’re built to spot seven main facial expressions: happy, sad, angry, fear, surprise, disgust, and neutral. We wanted anyone to be able to use it, so we made one smart change: every video frame gets resized before being processed. That one step sped things up, cut down on processor load, and kept our accuracy high. In the end, the system hums along at 25 frames per second, handling the messiness of the real world without breaking a sweat.

One piece really pops out—statistical tracking. The system doesn’t just spot an emotion and call it a day. It tracks how often each emotion shows up and maps out the emotional changes as a session goes on. Thanks to dynamic data structures, you get more than just quick snapshots—you see the whole story, how emotions shift over time. Bottom line? Pairing pre-trained VGG-Face models with this straightforward setup gets great results—even when faces disappear for a moment. The system rolls with whatever comes its way and just keeps going. It’s a solid base for real- time emotion analysis, and now researchers can actually track mood swings as they happen, or build interfaces that react right when you need them to. It’s a good step toward machines that actually get what people are feeling.

Keywords: - Face Detection, Emotion Recognition, Deep Learning, OpenCV, Artificial Intelligence.

Abstract: This paper presents a brain tumor detection system designed to assist early diagnosis using a multi-model machine learning approach. The system integrates MRI image analysis using Convolutional Neural Network (CNN) and symptom-based prediction using Random Forest. It combines both medical imaging and clinical data to improve accuracy and reliability. The system is implemented as a web-based application that allows users to upload MRI images or enter symptoms for preliminary screening. It targets healthcare support by providing fast, accessible, and effective tumor detection.

Keywords: Brain Tumor Detection, Machine Learning, Deep Learning, Convolutional Neural Network, Random Forest, MRI, Medical Image Analysis, Web Application.

Abstract: The increasing pace of diffusion of Artificial Intelligence (AI) and digitalization is radically re-organizing labour markets world-over, but micro-level empirical data on the interaction of automation exposure and organizational reskilling to influence employee job-security perceptions are limited. This paper is based on the Skill-Biased Technological Change (SBTC) theory and Human Capital Theory (HCT) and creates a mediated moderation model where organizational reskilling provision mediates and moderates the association between AI exposure and job-security concern. A total of 240 respondents in the manufacturing, logistics, retail, healthcare, and IT industries were used to collect primary data using a structured questionnaire. Pearson Chi-square, correlation analysis, descriptive statistics and multiple regression (OLS) were used. The chi-square test also ensured that there was no direct significant relationship between automation exposure and job-security concern (χ² = 3.28, df = 8, p = 0.916). But the regression analysis indicated that the reskilling provision has strong and negative predictive result of job-security concern (β = -0.34, p < 0.001) and that AI exposure interacting with reskilling has a strong attenuative effect on concern (β = -0.21, p = 0.018). The perception of positive automation had a strong influence on the optimism regarding the creation of new jobs (β = 0.47, p < 0.001). These results build on the SBTC model by showing that the exposure to automation is not the determinant of psychological employment outcomes but the organizational human-capital investment reaction. The research provides evidence to inform policymakers to build reskilling infrastructures and organization leaders to build human-AI partnerships.

Keywords: Artificial Intelligence; Digitalization; Workforce Transformation; Skill-Biased Technological Change; Human Capital Theory; Job-Security; Reskilling; Mediated Moderation; Regression Analysis