International Journal of Computational Learning & Intelligence

Explainable Machine Learning Models For Detecting Malware in PDF Files

B Chennakasava Reddy — 2025-04-20

The Portable Document Format (PDF) is widely used for document sharing, making it a common target for cyber threats. Attackers often embed malicious code within PDFs to exploit system vulnerabilities. Traditional malware detection techniques struggle to keep up due to evolving attack methods and reliance on predefined feature sets. This study presents an improved approach for detecting PDF malware using machine learning and explainability analysis. A comprehensive dataset of 15,958 PDF samples—comprising benign, malicious, and evasive files—is developed for this research. Three widely used PDF analysis tools (PDFiD, PDFINFO, and PDF-PARSER) are employed to extract meaningful features, alongside additional derived features that enhance classification accuracy. Through systematic feature selection and empirical evaluation, an optimal feature set is identified. The proposed method is tested with various machine learning classifiers, with the Random Forest model achieving approximately 2% higher accuracy compared to baseline models. Additionally, a decision tree is generated to enhance model interpretability, offering insights into classification rules. A comparative analysis with existing studies highlights key findings and advancements in PDF malware detection

Exploring Web Security Vulnerabilities Considering Man in the Middle and Session Hijacking

Shaik Faqrunnisa — 2025-04-15

Cybersecurity threats such as Man-in-the-Middle (MITM) attacks and Session Hijacking (SH) account for over 35% of web-based cyber intrusions, causing financial losses exceeding $6 billion annually. Despite extensive research on these attacks independently, a unified analysis remains underexplored. This study bridges that gap by conducting a Systematic Literature Review (SLR) on over 150 research papers from IEEE, ACM, and ScienceDirect, comparing MITM and SH in terms of attack frequency, methodologies, vulnerabilities, and countermeasures. Our findings indicate that MITM attacks constitute 27% of credential theft incidents, exploiting weak HTTPS encryption, phony server links, and packet sniffing. In contrast, Session Hijacking is responsible for 18% of unauthorized access cases, often leveraging TCP/UDP hijacking, cookie theft, and replay attacks. The study also reveals that 70% of successful MITM and SH attacks stem from improper session security configurations. To mitigate these risks, we propose an advanced cybersecurity framework integrating real-time behavioral analytics to detect anomalies with an 85% accuracy rate, significantly reducing unauthorized access attempts. By implementing adaptive security measures and AI-driven intrusion detection, organizations can enhance their defenses against these evolving threats

IoT - Enabled Machine learning for Ground Water Level monitoring in peatlands

P Parimala Kumari — 2025-04-20

Peatlands are a critical ecological concern due to their susceptibility to extensive carbon emissions during wildfires. Traditional methods for monitoring Ground Water Level (GWL) in these areas are labor-intensive, lack real-time insights, and impede proactive fire management. This study introduces an Internet of Things (IoT)-based system integrated with a neural network model for real-time GWL prediction. The proposed approach leverages atmospheric parameters to forecast GWL, allowing stakeholders to implement timely preventive measures to mitigate fire hazards. The neural network model exhibits high predictive accuracy, achieving a Root Mean Square Error (RMSE) ranging from 3.554 to 4.920. This ensures a 99% accuracy level within a deviation of 14.760 mm from actual GWL measurements. The study highlights the effectiveness of IoT-based solutions in overcoming the limitations of conventional GWL monitoring. By integrating neural networks with real-time data acquisition, the proposed framework offers a novel method for predicting GWL in resource-constrained regions.

Enhancing Predictive Maintenance in Smart Agriculture using Explainable Artificial Intelligence

T Prathima Reddy — 2025-04-20

The integration of Artificial Intelligence (AI) in Smart Agricultural Facilities (SAF) enhances efficiency but often lacks transparency, limiting its adoption by farmers. This study introduces a Predictive Maintenance (PdM) model powered by Explainable Artificial Intelligence (XAI) to improve both predictive accuracy and interpretability. The proposed model offers explanations across four key dimensions: data, model, outcome, and end-user, ensuring better understanding and usability for stakeholders. Experimental results demonstrate that the Long Short-Term Memory (LSTM) classifierimproves accuracy by 5.81%, while the eXtreme Gradient Boosting (XGBoost) classifier achieves a 7.09% increase in F1 score, 10.66% higher accuracy, and a 4.29% improvement in ROC-AUC. These enhancements lead to more precise maintenance predictions, reducing costs and improving reliability in SAF. Additionally, this study highlights data integrity, global and local model explanations, and counterfactual reasoning to enhance transparency in AI-driven PdM. By emphasizing interpretability beyond conventional accuracy metrics, this research contributes to advancing trustworthy AI applications in agriculture. Future research should explore multi-modal data integration and Human-in-the-Loop (HITL) systems to address ethical concerns such as Fairness, Accountability, and Transparency (FAT) in AI-driven agricultural technologies.

Identification of Visual Learners Using Raw EEG

P Arshiya Khannam — 2025-04-20

The project titled "IDENTIFICATION OF VISUAL LEARNERS USING RAW ELECTROENCEPHLOGRAPHY" addresses the challenge of accurately identifying visual learners, who are a significant portion of the student population that benefits from visual stimuli in their learning processes. Traditional methods of identifying learning styles, such as self-report questionnaires, are often subjective and prone to biases, highlighting the need for more objective approaches. To tackle this issue, the project employs a novel hybrid methodology that integrates Convolutional Neural Networks (CNNs) and Long Short-Term Memory networks (LSTMs) with a Random Forest classifier. This approach leverages the strengths of CNNs in extracting spatial features from raw EEG data, while LSTMs capture the temporal dependencies inherent in the sequential nature of EEG signals. The implications for educational practices are profound. This project not only paves the way for personalized educational strategies tailored to individual learning styles but also emphasizes the potential of neuroeducational techniques in enhancing learning outcomes. By utilizing advanced machine learning algorithms, educators can develop targeted interventions that align with students' cognitive preferences, ultimately optimizing the learning experience and fostering better academic performance.

Blockchain Technology: A Catalyst For Eco-Friendly Product Validation

K Satya Mounika — 2025-04-20

The rise of blockchain technology has reshaped competition between traditional and ecofriendly products by offering a means to verify sustainability claims. This study uses game theory to examine how blockchain integration influences market dynamics between retailers selling eco-friendly and conventional goods. Two pricing models are considered: one for a firm using blockchain certification for its eco-friendly product and another for a manufacturer of traditional products. The study assesses two key factors blockchain adoption costs and product quality choices. Findings indicate that blockchain does not inherently expand the market for eco-friendly products but intensifies competition as eco-conscious consumer numbers grow. While blockchain may alleviate some competitive pressures, its presence alone does not guarantee an edge for eco-friendly products. Companies leveraging blockchain must also secure strong bargaining power to outperform conventional products.

Machine Learning For Medicare Fraud Detection: Tackling Class Imbalance With SMOTE-ENN

A Krishnapriya — 2025-04-20

The realm of healthcare fraud detection is continually changing and encounters substantial obstacles, especially when dealing with data imbalance problems. Earlier research primarily concentrated on standard machine learning (ML) methods, which often have difficulty with imbalanced data. This issue manifests in several ways. It involves the danger of overfitting with Random Oversampling (ROS), the creation of noise by the Synthetic Minority Oversampling Technique (SMOTE), and the possible loss of vital information with Random Undersampling (RUS). Furthermore, enhancing model performance, examining hybrid resampling techniques, and refining evaluation metrics are essential for achieving greater accuracy with imbalanced datasets. In this study, we introduce a new technique to address the problem of imbalanced datasets in healthcare fraud detection, specifically focusing on the Medicare Part B dataset. Initially, we carefully remove the categorical feature ‘‘Provider Type’’ from the dataset. This enables us to create new, synthetic instances by randomly copying existing types, thus increasing the diversity within the minority class. Subsequently, we implement a hybrid resampling method called SMOTE ENN, which combines the Synthetic Minority Over-sampling Technique (SMOTE) with Edited Nearest Neighbours (ENN).

Optimizing Spam Filtering on the Social Web of Things with Supervised Sampling Methods

Charitha — 2025-04-20

The rise of digital communication has led to an increasing challenge in detecting and filtering spam messages, which negatively affect user experience and system performance. Conventional spam detection methods often struggle with imbalanced datasets, reducing their classification effectiveness. This study presents an innovative supervised learning model that integrates Negative Selection Density Clustering with Down sampling (NSDC-DS) and a Naïve Bayes Support Vector Machine (NBSVM) to enhance spam detection accuracy. NSDC-DS improves data balance by clustering based on density similarity, ensuring better representation of minority classes. Additionally, Principal Component Analysis with Stochastic Gradient Descent (PCA-SGD) is employed to optimize feature selection and enhance model performance. Experimental analysis on diverse communication datasets demonstrates that the proposed approach surpasses traditional classifiers in both accuracy and efficiency. The findings confirm that this method offers a reliable and optimized solution for detecting spam messages in online communication platforms.

Intelligent Threat Detection in CPS-IoT Networks Using A Hybrid CNN-DBN Model with Saeho Optimization

B Mamatha — 2025-04-20

The Internet of Things (IoT) is integral to smart cities and diverse societal applications, yet its large-scale implementation is hindered by significant security vulnerabilities and cyber threats. Conventional security measures frequently struggle to tackle the distinct challenges associated with IoT-driven cyber-physical systems, highlighting the need for advanced techniques like Deep Learning (DL) for robust anomaly detection. This research introduces an innovative framework that utilizes a hybrid classification strategy by combining a Deep Belief Network (DBN) with a Convolutional Neural Network (CNN). To enhance detection accuracy, the framework incorporates an innovative optimization technique called Seagull Adapted Elephant Herding Optimization (SAEHO). The "Hybrid Classifier + SAEHO" model processes extracted features from network traffic data, effectively distinguishing between malicious and benign activity. Experimental evaluations on two datasets demonstrate superior performance in terms of sensitivity, precision, accuracy, and specificity when compared to conventional methods. These results highlight the model’s potential in fortifying IoT security and offering a reliable mechanism for mitigating cyber threats in real-world applications.

An Innovative Method For Ensuring The Accuracy Of Online Exam Results Via Blockchain Technology

M Sireesha — 2025-04-17

he rapid adoption of Learning Management Systems (LMS) has revolutionized education, particularly in online assessments. However, traditional exam management systems rely on centralized databases, making them vulnerable to security threats such as hacking, unauthorized access, and result manipulation. This research proposes a blockchain-based framework to enhance the security, transparency, and reliability of online exam results. By leveraging blockchain’s decentralized nature, cryptographic security, and proof-of-stake validation, the proposed system ensures tamper-proof record-keeping. The framework is integrated with Moodle LMS, enabling seamless and secure examination administration. Comparative analysis with conventional systems demonstrates that blockchain technology significantly improves exam security, mitigates data tampering risks, and provides an immutable audit trail. The findings confirm that blockchain-based exam management ensures academic integrity and enhances trust in online assessments.

Risk Prediction in Software Engineering: A Multi-Class Based Approach with FEPP

M Venkata Ramana — 2025-04-20

Accurate risk prediction plays a crucial role in the successful execution of software projects by identifying potential threats and weaknesses early in the development process. This paper introduces an innovative multi-class, role-specific strategy for predicting risks in software engineering, incorporating Feature Extraction and Prioritization Paradigm (FEPP) to improve the precision and efficiency of risk detection. The model is specifically tailored to handle complex risk patterns associated with various team roles such as developers, testers, and project managers. Through this methodology, the system aims to support proactive risk mitigation, ensuring better project outcomes.

Distributed Transformer Framework for Financial Anomaly Detection

S Mahinoor Begum — 2025-04-15

Financial fraud is a significant concern for investors and financial institutions, leading to substantial economic losses. Traditional fraud detection techniques often struggle with challenges such as low accuracy, slow processing times, and limited adaptability across different financial sectors. To address these issues, this paper introduces a distributed knowledge distillation framework utilizing Transformer models. The approach employs a multi-attention mechanism to highlight important features, followed by a feed-forward neural network for extracting high-level representations. A final neural network classifier then determines fraudulent activity. Additionally, to tackle inconsistencies in financial data and imbalanced distributions across industries, a distributed knowledge distillation algorithm is proposed. Financial fraud cases causing serious damage to the interests of investors are not uncommon. As a result, a wide range of intelligent detection techniques are put forth to support financial institutions’ decision-making. Currently, existing methods have problems such as poor detection accuracy, slow inference speed, and weak generalization ability. Therefore, we suggest a distributed knowledge distillation architecture for financial fraud detection based on Transformer. Firstly, the multi-attention mechanism is used to give weights to the features, followed by feed-forward neural networks to extract high-level features that include relevant information, and finally neural networks are used to categorize financial fraud. Secondly, for the problem of inconsistent financial data indicators and unbalanced data distribution focused on different industries, a distributed knowledge distillation algorithm is proposed. This algorithm combines the detection knowledge of the multi-teacher network and migrates the knowledge to the student network, which detects the financial data of different industries. This method integrates insights from multiple teacher models and transfers their knowledge to a student network, enhancing fraud detection capabilities across diverse industries. Experimental evaluations demonstrate that the proposed approach surpasses traditional methods, achieving an F1 score of 92.87%, accuracy of 98.98%, precision of 81.48%, recall of 95.45%, and an AUC score of 96.73%.

Innovative Data Science Model for Analysis on Pesticide Poisoning Using Supervised Learning

M Venkata Ramana — 2025-04-20

In a Data Science project, assessing data relevance and identifying patterns that support decision-making based on domain-specific knowledge are critical. Additionally, establishing clear methodologies and comprehensive documentation is essential to guide the project from its initial stages to completion. This study introduces a structured Data Science model, covering the entire process from data collection to model training, aimed at enhancing knowledge discovery. The motivation behind this model stems from limitations in existing Data Science methodologies, particularly the absence of practical, step-by-step guidance for data preparation and deployment. The proposed model, called "Data Refinement Cycle with Supervised Machine Learning (DRC–SML)," was specifically designed to assist healthcare professionals in diagnosing pesticide poisoning among rural workers. The dataset for this project, based on scientific research, included 1027 samples containing toxicity biomarker data and clinical analyses. The model achieved an impressive 99.62% accuracy with only 28 decision rules, significantly improving healthcare practices and quality of life in rural areas. The results validate the effectiveness of the DRC–SML model, demonstrating its potential for enhancing predictive analytics in healthcare and other domains.

A Multi-Layer Trust Framework for Self-Sovereign Identity on Blockchain

M Jyothi — 2025-04-20

The growing sophistication of deepfake technology poses a significant challenge to remote identity verification systems, particularly in electronic Know Your Customer (eKYC) applications. Many existing deepfake detection datasets lack the necessary features to assess eKYC systems effectively, as they do not include essential factors like head movements and facial verification protocols. To address this gap, we introduce eKYC-DF, a large-scale dataset comprising over 228,000 high-quality synthetic and real videos, representing diverse demographics. This dataset is designed to facilitate the development and evaluation of eKYC systems by incorporating various head poses, facial expressions, and verification benchmarks. Additionally, our dataset provides protocols for both deepfake detection and facial recognition assessments, making it a valuable resource for enhancing identity-proofing security. The eKYC-DF dataset, along with evaluation tools and pre-trained models, is publicly available to researchers for further study and development.

AI-Driven Emotion Analytics For Emergency Management in Tourism Using Improved CNN

P Chandra Obul Reddy — 2025-04-20

Emotion recognition plays a critical role in enhancing human-computer interactions, particularly in dynamic environments like the tourism industry. During emergency events, understanding tourists' emotions can aid in decision-making, safety measures, and overall experience management. This study leverages deep learning methodologies, particularly Convolutional Neural Networks (CNN), to classify and analyze emotional states. The proposed system integrates image preprocessing, feature extraction, and advanced classification techniques to improve accuracy and efficiency. By incorporating real-time emotion detection, the model enhances responsive management strategies, ensuring improved safety and customer satisfaction.

Predicting EV Battery Lifespan Using Machine Learning

N Vasavi — 2025-04-20

The continuous advancement of electric vehicle (EV) technology has heightened the emphasis on sustainable energy storage, making lithium-ion batteries a crucial component. Ensuring battery reliability and longevity is essential for optimizing EV performance and reducing maintenance costs. This study explores the prediction of Remaining Useful Life (RUL) for lithium-ion batteries using advanced Machine Learning (ML) models, specifically Random Forest (RF) and Support Vector Machine (SVM). Accurate RUL estimation enhances battery management, prevents failures, and improves safety.A comprehensive dataset from the NASA Ames Prognostics Center of Excellence is preprocessed, with the One-way ANOVA method applied for optimal feature selection. Data normalization techniques are employed to enhance model consistency, while hyperparameter tuning (HPT) optimizes predictive performance. Real-time factors such as temperature fluctuations and usage cycles are incorporated to analyze their impact on battery degradation. The proposed system provides deeper insights into battery aging trends, enabling proactive maintenance strategies.Model performance is evaluated using R2 score and Mean Squared Error (MSE), where the RF model achieves an R2 score of 0.83 and an MSE of 1.67, demonstrating high reliability. The results contribute to improving battery efficiency and safety through predictive modeling, facilitating better battery management in EVs. By leveraging ML-driven predictive analytics, this research supports the advancement of sustainable and cost-effective energy solutions, promoting wider EV adoption and a greener future.

Emotion Recognition Using Multi-Scale Auto-Encoders with Cross Session Adoption

G ChennaKesava Reddy — 2025-04-20

Emotion recognition from EEG (electroencephalography) signals is a challenging yet promising area of research, with applications ranging from mental health monitoring to adaptive human-computer interactions. Traditional approaches, such as those using Random Forest algorithms, have shown potential but often fall short in effectively capturing the complex temporal and spatial patterns inherent in EEG data. In this study, we propose a novel framework employing Multi-Scale Masked Autoencoders (MSMAE) combined with Convolutional Neural Networks (CNNs) for cross-session emotion recognition. Utilizing the Seed IV EEG dataset, our method leverages the multi-scale feature extraction capabilities of MSMAE to handle varying signal frequencies and the powerful pattern recognition abilities of CNNs to enhance classification accuracy. The MSMAE framework pre-trains the CNN by reconstructing the masked EEG signals at different scales, enabling it to learn robust and generalized features across different sessions. Comparative evaluations demonstrate that our proposed MSMAE-CNN model significantly outperforms the existing Random Forest algorithm, providing a more reliable and effective solution for emotion recognition in diverse and dynamic environments. This advancement not only highlights the potential of deep learning models in EEG-based emotion recognition but also sets a new benchmark for future research in this field

AI-Driven Advanced Techniques for Detecting Dry Eye Disease Using Multi-Source Evidence: Case Studies, Applications, Challenges, and Future Perspectives

P Parimala kumari — 2025-04-20

This study examines the transformative potential of Artificial Intelligence (AI) in the early diagnosis and prognostic evaluation of Dry Eye Disease (DED), aiming to elevate the precision of clinical interventions for eye-care specialists. Despite AI’s promising capabilities, its deployment is hindered by challenges such as diverse diagnostic inputs, the multifaceted etiology of DED, and the integration of cross-disciplinary expertise, all of which affect the transparency, reliability, and practical utility of AI-driven detection systems. Through a thorough analysis of the past five years, we assess datasets, diagnostic criteria, standardized benchmarks, and cutting-edge AI algorithms central to DED detection. We organize DED diagnostic strategies into three categories based on their alignment with AI technologies: (1) methods rooted in established benchmarks or comparable standards, (2) pioneering AI techniques with distinct advantages, and (3) supportive approaches enhancing AI-based detection. This research proposes refined diagnostic protocols, promotes the synthesis of multiple evidence sources, and delineates future research directions to guide subsequent investigations. By elucidating foundational insights, innovative methodologies, persistent challenges, and prospective pathways, this work advances ophthalmic disease detection, highlighting AI’s critical role in both scholarly inquiry and clinical ophthalmology.