Daily Archives: June 27, 2026

Authors: Atharv Nitin Gore, Rushikesh Vijay Kolhe, Samarth Suresh Gaikwad, Professor Snehal Phate

Abstract: This paper proposes an AI-powered lead management platform designed to optimize sales pipeline efficiency through a Hybrid Machine Learning Classifier. The system incorporates multi-channel lead capture, XGBoost-based lead scoring, BERT-driven intent detection, lead deduplication, and CRM synchronization to enable real-time qualification and conversion of leads. With an F1 Score of 87% and AUC-ROC of 93%, it ensures a reliable and data-driven pipeline management experience. The system supports various lead sources including web forms, social media, email, and API integrations, making it highly adaptable for B2B and B2C enterprises. Ethical considerations are addressed through strong privacy safeguards, JWT-based authentication, and GDPR-compliant data management. Additionally, it minimizes manual sales effort, reduces lead response time by 89%, and enhances conversion rates by 79%. This solution establishes a reliable framework for secure, automated, and scalable lead management in digital marketing and sales operations. By leveraging advanced AI techniques such as XGBoost scoring, BERT intent detection, and fuzzy deduplication, the system effectively prioritizes high-value prospects to maximize pipeline conversion.

DOI: https://doi.org/10.5281/zenodo.20962801

Published by: vikaspatanker

Authors: Miss. Pemmanaboyina Durga Devi, Miss. Savarapu Suhasini

Abstract: Cloud computing has become the foundation of modern digital services by providing scalable, flexible, and cost-effective computing resources for organizations across various domains. Despite its widespread adoption, the increasing complexity of cloud infrastructures has introduced numerous security challenges, including unauthorized access, insecure configurations, application vulnerabilities, distributed denial-of-service (DDoS) attacks, and abnormal network activities. Conventional cloud security mechanisms primarily rely on rule-based detection techniques, which often struggle to identify sophisticated and previously unknown cyber threats in dynamic cloud environments. To overcome these limitations, this paper proposes an intelligent machine learning-based framework for cloud vulnerability detection and threat prevention. The proposed framework analyzes security-related information collected from system logs, network traffic records, cloud service activities, and vulnerability reports to identify malicious behavior and potential security risks. Comprehensive data preprocessing and feature engineering techniques are employed to improve data quality before training multiple machine learning models, including Decision Tree, Support Vector Machine (SVM), Random Forest, K-Nearest Neighbors (KNN), and Isolation Forest. The effectiveness of these algorithms is evaluated using performance metrics such as accuracy, precision, recall, F1-score, confusion matrix, and ROC-AUC analysis. Experimental results demonstrate that the Random Forest model achieves superior detection performance by accurately identifying cloud vulnerabilities while maintaining a low false alarm rate. The proposed framework enables real-time threat monitoring, intelligent anomaly detection, and adaptive security analysis, thereby improving the resilience, reliability, and overall protection of distributed cloud infrastructures against evolving cyber threats.

DOI: http://doi.org/10.61137/ijsret.vol.12.issue3.479

Published by: vikaspatanker

Authors: Miss. Tatipaka Pooja, Miss. Savarapu Suhasini

Abstract: Accurate credit risk assessment is essential for financial institutions to minimize loan defaults and support effective lending decisions. Conventional loan evaluation processes largely depend on manual analysis of customer financial information, making them time-consuming, inconsistent, and susceptible to human bias. With the rapid advancement of machine learning, intelligent prediction models have emerged as efficient solutions for automating credit risk evaluation and improving decision-making accuracy. This paper presents an intelligent credit risk prediction framework that utilizes machine learning algorithms to classify loan applicants based on their probability of loan repayment or default. The proposed framework analyzes customer financial and demographic attributes, including credit history, checking account status, employment status, loan amount, loan duration, and applicant age. Data preprocessing techniques such as missing value handling, outlier removal, categorical feature encoding, and feature scaling are employed to enhance data quality before model training. Multiple machine learning algorithms, including Decision Tree, Random Forest, Support Vector Machine (SVM), Naïve Bayes, Multi-Layer Perceptron (MLP), and a Stacking Ensemble model, are implemented and comparatively evaluated using performance metrics such as accuracy, precision, recall, F1-score, confusion matrix, and ROC-AUC analysis. Experimental results indicate that the ensemble learning approach consistently outperforms individual classifiers by achieving higher prediction accuracy and improved generalization capability. The proposed framework provides a reliable, scalable, and data-driven solution for intelligent credit risk assessment, enabling financial institutions to improve loan approval decisions, reduce financial losses, and strengthen overall credit risk management.

DOI: http://doi.org/10.61137/ijsret.vol.12.issue3.478

Published by: vikaspatanker

Authors: Mr. Chokkakula Chaitanya, Miss. Savarapu Suhasini

Abstract: Heart disease is one of the leading causes of mortality worldwide, making early and accurate diagnosis essential. This study proposes a next-generation heart disease prediction framework using Quantum Machine Learning (QML) and presents a comparative evaluation with traditional machine learning approaches. A clinical heart disease dataset containing attributes such as age, gender, blood pressure, cholesterol level, and heart rate is pre-processed, balanced, and divided into training and testing sets. Traditional algorithms, including Logistic Regression, Support Vector Machine (SVM), Naïve Bayes, Decision Tree, K-Nearest Neighbors (KNN), and Linear Discriminant Analysis (LDA), are compared with Quantum Machine Learning models for disease prediction. The models are evaluated using 5-fold cross-validation with accuracy, precision, recall, F1-score, and ROC-AUC as performance metrics. Results show that Logistic Regression and Linear Discriminant Analysis achieve the best performance among classical models, while Quantum Machine Learning demonstrates competitive prediction capability with improved feature representation. The proposed framework highlights the potential of QML as a scalable and intelligent solution for next-generation heart disease prediction and clinical decision support.

DOI: http://doi.org/10.61137/ijsret.vol.12.issue3.477

Published by: vikaspatanker

Authors: Mr. Boddu Pavan Kumar, Miss. Savarapu Suhasini

Abstract: The increasing demand for intelligent dietary assessment and nutrition monitoring has accelerated research in automated food image classification systems. However, accurately identifying food categories remains challenging due to significant variations in appearance, illumination, background complexity, and similarities among visually related food items. Conventional machine learning techniques, which rely on handcrafted feature extraction, often fail to capture the intricate visual characteristics required for robust food recognition. To overcome these limitations, this paper presents a hybrid food image classification framework that integrates transfer learning-based feature extraction with machine learning classifiers. Pre-trained deep learning architectures, including EfficientNet, DenseNet, and MobileNet, are employed to learn rich and discriminative visual representations from food images without requiring extensive model training. The extracted deep features are subsequently processed using advanced machine learning algorithms such as Random Forest and XGBoost to perform accurate food category prediction. This hybrid strategy effectively combines the representational strength of deep neural networks with the computational efficiency and interpretability of classical machine learning methods. Experimental evaluation demonstrates that the proposed framework achieves superior classification accuracy, precision, recall, and F1-score compared with conventional image classification approaches. Furthermore, the model exhibits improved robustness when handling diverse food images captured under varying environmental conditions. The proposed framework has significant potential for practical deployment in applications such as intelligent nutrition monitoring, automated calorie estimation, healthcare support systems, and smart dietary recommendation platforms, contributing to the development of reliable AI-driven food analysis solutions.

DOI: http://doi.org/10.61137/ijsret.vol.12.issue3.476

Published by: vikaspatanker

Authors: Mr. Suryaashokkumar Siriki, Miss. Savarapu Suhasini

Abstract: Rapid and accurate fire detection is essential for minimizing human casualties, reducing property damage, and enabling timely emergency response. Conventional fire detection systems primarily depend on smoke, heat, and gas sensors, which often experience delayed response, high false alarm rates, and limited effectiveness in complex or large-scale environments. Recent advances in deep learning and computer vision have enabled intelligent visual monitoring systems capable of identifying fire incidents directly from surveillance imagery. This paper presents a deep learning-based intelligent fire detection and early warning framework that employs Convolutional Neural Networks (CNNs) to automatically classify surveillance images into fire and non-fire categories. The proposed framework utilizes a comprehensive image preprocessing pipeline, including resizing, normalization, and data augmentation techniques such as rotation, scaling, zooming, and horizontal flipping to improve model robustness and generalization. Training optimization strategies, including Early Stopping and ReduceLROnPlateau, are incorporated to enhance learning stability and prevent overfitting. The performance of the proposed CNN model is compared with conventional machine learning algorithms, including Logistic Regression, K-Nearest Neighbors (KNN), and AdaBoost, using evaluation metrics such as accuracy, precision, recall, F1-score, confusion matrix, and ROC-AUC analysis. Experimental results demonstrate that the CNN-based framework achieves superior classification performance by effectively learning complex visual characteristics of flames and smoke while maintaining high detection accuracy and a low false alarm rate. The system further integrates an automated alert mechanism that instantly generates notifications upon fire detection, supporting rapid emergency intervention. The proposed framework provides an intelligent, scalable, and cost-effective solution for real-time fire monitoring and can be effectively deployed in smart buildings, industrial facilities, public infrastructures, and smart city surveillance systems to strengthen fire safety management and disaster prevention.

DOI: http://doi.org/10.61137/ijsret.vol.12.issue3.475

Published by: vikaspatanker

Authors: Miss. Kathula Lakshmi, Miss. Savarapu Suhasini

Abstract: The rapid adoption of Unmanned Aerial Vehicles (UAVs) in applications such as surveillance, precision agriculture, disaster response, logistics, and intelligent transportation has significantly increased the demand for secure and reliable communication networks. However, the wireless nature of UAV communication exposes these systems to a wide range of cyber threats, including GPS spoofing, data injection, denial-of-service (DoS), and network intrusion attacks, which can compromise mission integrity and operational safety. To address these security challenges, this paper presents a machine learning-based cyber attack detection framework for UAV communication networks. The proposed framework employs comprehensive data preprocessing, feature engineering, and intelligent classification techniques to analyze UAV telemetry data, communication signals, and operational parameters for identifying malicious activities. Multiple machine learning models are utilized to distinguish normal UAV behavior from cyber attack scenarios through behavioral pattern analysis and anomaly detection. The framework is evaluated using standard performance metrics, including accuracy, precision, recall, F1-score, and ROC-AUC, to assess its detection capability and reliability. Experimental results demonstrate that the proposed framework effectively detects various cyber attacks with high detection accuracy, low false positive rates, and efficient response time. By integrating intelligent machine learning algorithms into UAV cybersecurity, the proposed approach enhances communication security, improves system resilience, and supports the development of reliable and autonomous drone operations in dynamic network environments.

DOI: http://doi.org/10.61137/ijsret.vol.12.issue3.474

Published by: vikaspatanker

Authors: Rakesh Dondapati

Abstract: The rapid diffusion of generative AI tools and autonomous agents has generated a pervasive and largely ungoverned organizational phenomenon: shadow AI, whereby employees and teams deploy AI capabilities outside formal information technology governance and procurement processes. While shadow AI may generate local productivity improvements and serve as an incubator for grassroots innovation, it simultaneously exposes organizations to compounding risks across data security, regulatory compliance, intellectual property control, and operational integrity domains. This study investigates the dual character of shadow AI — as both an organizational threat and an innovation catalyst — and examines the conditions under which adaptive governance structures enable firms to convert unauthorized AI experimentation into sanctioned strategic capability. Drawing on a multi-source dataset comprising IT leader survey responses, employee-level AI usage telemetry, security incident reports, patent disclosures, and longitudinal firm performance data from 487 firms across seven industry sectors (2022–2026), the study develops and validates the Shadow AI Prevalence Index (SAPI) and the Governance Adaptiveness Score (GAS). Structural equation models demonstrate that SAPI is positively associated with risk exposure (β = 0.48, p < .001) but that governance adaptiveness significantly moderates this relationship (interaction β = –0.27, p < .001), and independently predicts innovation output (β = 0.41, p < .001) and organizational resilience (β = 0.48, p < .001). Six inductively derived qualitative themes from 48 executive interviews illuminate the mechanisms linking governance adaptiveness to shadow AI outcomes. The study advances a theory of adaptive AI governance, provides the first large-scale empirical examination of the shadow AI prevalence-performance relationship, and delivers a practical Shadow-to-Sanctioned AI conversion framework for enterprise practitioners. Findings indicate that the critical governance imperative is not the elimination of shadow AI — which is both practically infeasible and strategically self-defeating — but its structured transformation from hidden organizational risk into visible competitive capability.

DOI: http://doi.org/10.5281/zenodo.20958484

Published by: vikaspatanker

Authors: Mr. Appalla Yazna Surya Sai Kiran, Miss. Savarapu Suhasini

Abstract: The rapid growth of social media platforms and online communication has significantly increased the volume of user-generated content, creating new challenges in identifying toxic language, hate speech, cyberbullying, and abusive comments. These harmful interactions negatively affect online communities, user well-being, and digital safety, highlighting the need for intelligent and automated content moderation systems. This paper presents ToxiShield, a next-generation intelligent framework for toxic comment detection that integrates Machine Learning (ML) and Natural Language Processing (NLP) techniques to accurately classify online comments as toxic or non-toxic. The proposed framework employs comprehensive text preprocessing, including tokenization, stop-word removal, text normalization, lemmatization, and feature extraction using Term Frequency–Inverse Document Frequency (TF-IDF) and word embedding techniques to generate meaningful textual representations. To evaluate the effectiveness of the proposed framework, multiple classification algorithms, including Naïve Bayes, Logistic Regression, Support Vector Machine (SVM), Random Forest, and Convolutional Neural Networks (CNN), are implemented and comparatively analysed using performance metrics such as accuracy, precision, recall, and F1-score. Experimental results demonstrate that deep learning-based models, particularly CNN, achieve superior performance in identifying complex contextual toxicity patterns compared with traditional machine learning methods. The proposed ToxiShield framework provides an efficient, scalable, and intelligent solution for automated online content moderation, contributing to safer digital communication environments and promoting respectful interactions across social media platforms and online communities.

DOI: http://doi.org/10.61137/ijsret.vol.12.issue3.472

Published by: vikaspatanker

Authors: Professor Mayuri Dongre, Saurabh Bhoyar, Sanskar Karnewar

Abstract: Fraudulent voice calls have become a prominent cyber threat in the contemporary telecommunication environment as the usage of online banking, UPI transactions, mobile wallets, and instant messaging services becomes widespread. The perpetrators of cybercrime resort to fraudulent activities such as voice calls, phishing attacks, OTP manipulation, lottery scams, insurance scams, loan scams, and identity deception. The consequences include substantial monetary damage and grave security vulnerabilities. Existing techniques for spam detection in phone calls depend upon manual reporting, blacklisting, and basic rules-based filtering algorithms. However, these methods prove ineffective against newly emerging and evolving forms of fraud, particularly when the perpetrator changes their phone number and employs advanced social engineering techniques. Therefore, there is a need to develop an efficient and automated fraud detection system. In this paper, we propose a machine learning-based method to detect and analyze fraudulent phone calls. Using the following indicators for call behavior analysis, duration of a call, frequency of calls, suspicious phrases, time of calls, and voice pattern recognition, our approach is intended to identify and classify every call as either fraudulent or legitimate. For better prediction and detection, such machine learning models as Naive Bayes, Logistic Regression, Random Forest, and Support Vector Machine (SVM) will be applied.

DOI: https://doi.org/10.5281/zenodo.20956340

Published by: vikaspatanker