PRATHMESH GUNJGUR

Scopus Publications

Hybrid Reinforcement Learning-Based Anomaly Detection in AWS CloudTrail Logs
Prathmesh N. Gunjgur, Sudhir N. Dhage
2025 IEEE International Conference on Computer Electronics Electrical Engineering and their Applications Ic2e3 2025, 2025
Cloud security threats are becoming increasingly sophisticated, necessitating proactive and adaptive anomaly detection systems. This paper presents HRL-AD (Hybrid Reinforcement Learning-Based Anomaly Detection), an AI-driven framework designed for real-time threat detection in AWS CloudTrail logs. The proposed model integrates multiple machine learning techniques: unsupervised learning (Autoencoders, Isolation Forest) to detect subtle anomalies, supervised classification (XGBoost, Random Forest) to enhance accuracy, and Deep Q-Learning (DQN) to dynamically adjust detection thresholds based on feedback. This hybrid approach significantly improves anomaly classification by continuously learning from cloud behavior patterns. Experimental results demonstrate that HRL-AD achieves a Silhouette Score of 0.85, Precision of 0.92, Recall of 0.89, and F1-Score of 0.91, outperforming traditional models in both accuracy and false positive reduction. The study underscores the potential of reinforcement learning in cloud security and paves the way for future research in multi-cloud anomaly detection, adversarial robustness, and explainable AI, offering scalable and adaptive protection for modern cloud environments.
A Self-Driving Car Platform Using Raspberry Pi and Arduino
Vikrant Shahane, Hrushikesh Jadhav, Mihir Sansare, Prathmesh Gunjgur
2022 6th International Conference on Computing Communication Control and Automation Iccubea 2022, 2022
Cars that can drive themselves have long been the stuff of science fiction. However, this fiction will become a reality thanks to the self-driving automobile within the next several years. Self-driving cars are vehicles that navigate to a destination without the assistance of a human. Numerous prominent firms and developers have invested heavily in this field, creating their own self driving car systems. The fascinating topic of self-driving cars served as the inspiration for this study, which aims to develop a self-driving platform. This paper proposes a working model of a Self Driving car using Raspberry Pi, Arduino Uno and a camera based approach. The three major modules that the car uses to perform are lane detection, obstacle detection, and traffic sign detection. The camera module, which is installed on the car's roof, captures the live stream images and sends them to the Raspberry Pi, which processes them and sends them to all three modules. Algorithms such as Canny Edge Detection and the Hough Transform are utilised for Lane Detection. Based on the results of these algorithms, the car predicts the direction it wants to move in. The Traffic Sign Detection Module identifies traffic signs using a CNN and OpenCV-based approach. Obstacle Detection uses the HAAR Cascade technique to detect objects that may be encountered on the road, such as cars and pedestrians. These modules integrated onto the self-driving car moves it autonomously with maximum accuracy ranging from 95 to 97%.
A Recommendation System for Integrated Agriculture Using Neural Networks with Random Forest Algorithm
Shubham Gaud, Rugveda Sarambale, Shreya Kale, Prathmesh Gunjgur
2022 6th International Conference on Computing Communication Control and Automation Iccubea 2022, 2022
Agriculture and its affiliated sectors are significant sources of income in rural India. However, the agricultural yield per hectare is poor compared to worldwide norms. A difficulty experienced by the farmers is that the suitable crop is not selected for their soil conditions, resulting in a diminished yield. This study solves this difficulty by proposing a recommendation system with an accuracy of 96% that efficiently uses a random forest algorithm to select a crop based on site-specific variables. Plant diseases affect their unique species' growth; consequently, early detection is vital. Therefore to categorize and detect symptoms of plant illnesses, Convolutional Neural Network (CNN) architecture with an accuracy of 75% is utilized with numerous visualization techniques. Early diagnosis would aid in minimizing pesticide usage; further CNN model for the guiding of pesticides is also adopted with an overall accuracy of 81%. In addition, these strategies are evaluated using a range of performance metrics. Agriculture is an underdevelopment field lacking a solid trade mechanism to sell the harvest, resulting in farmers' benefit. An E-commerce platform has been built where farmers can sell daily generated commodities straight from their field location or their residences without adding the market—incorporating Artificial intelligence into agriculture methods has proven helpful in modernizing farming in a suggested system comprising crop, fertilizer, pesticide recommendation, and plant disease detection.
A Robust Captcha Scheme for Web Security
Yash Raut, Shreyash Pote, Harshank Boricha, Prathmesh Gunjgur
2022 6th International Conference on Computing Communication Control and Automation Iccubea 2022, 2022
The internet has grown increasingly important in everyone's everyday lives due to the availability of numerous web services such as email, cloud storage, video streaming, music streaming, and search engines. On the other hand, attacks by computer programmes such as bots are a common hazard to these internet services. Captcha is a computer program that helps a server-side company determine whether or not a real user is requesting access. Captcha is a security feature that prevents unauthorised access to a user's account by protecting restricted areas from automated programmes, bots, or hackers. Many websites utilise Captcha to prevent spam and other hazardous assaults when visitors log in. However, in recent years, the complexity of Captcha solving has become difficult for humans too, making it less user friendly. To solve this, we propose creating a Captcha that is both simple and engaging for people while also robust enough to protect sensitive data from bots and hackers on the internet. The suggested captcha scheme employs animated artifacts, rotation, and variable fonts as resistance techniques. The proposed captcha technique proves successful against OCR bots with less than 15% accuracy while being easier to solve for human users with more than 98% accuracy.
SISA: A secret-sharing scheme application for cloud environment
Rohit B. Chilwant, Taral S. Sarvagod, Kunal R. Kumbhar, Prathmesh N. Gunjgur, Amarsinh V. Vidhate
Proceedings of the 4th International Conference on Communication and Electronics Systems Icces 2019, 2019
Security defies one of the galactic barrier when pondering the use of cloud services. Data outsourcing in the cloud (DOC) is conventionally based on data encryption, which imparts impregnable security, but deteriorating the efficiency and rimming the functionalities of cloud. In this proposal we aim to possess them by proposing a secret sharing scheme that depends on partitioning the data and distributed storage over multi-clouds. The scheme used in the proposal uses a combination of symmetric (AES) and asymmetric (RSA) encryption techniques to share the data among the peers in the cloud environment. The proficiency of the suggested scheme has been demonstrated by the empirical results.
Semantic Analysis of Wikipedia documents using Ontology
Prachi Banik, Sonali Gaikwad, Anagha Awate, Shahabaj Shaikh, Prathmesh Gunjgur, Puja Padiya
2018 IEEE International Conference on System Computation Automation and Networking Icsca 2018, 2018
There is a boom in the growth of information available freely on the web where a search engine builds for a decisive component in understanding the content of the web pages and also serving the user queries according to their relevant information. The semantic web offers a hopeful approach in this context, ontologies can semantically seize concepts for any issue which will empower tools to accord the data semantically. In this paper, a proposed technique is developed which uses a score or weight based semantic relation between the user queries and gives a more relevant result. This system is moderated to Wikipedia related article as they are extracted from Wikipedia api. The similarity level between two articles is computed based on keyword content by computing similarity between two documents. We study various proposal in this regard thus the proposed system tries to optimize the results and the state-of-the-art analysis is presented. Likened to other similarity method, the proposed technique shows the highest Pearson correlation coefficient.
Session rate prediction for multimedia streaming
Prathmesh N. Gunjgur, Amarsinh V. Vidhate
Souvenir of the 2014 IEEE International Advance Computing Conference Iacc 2014, 2014
Uses of multimedia on video and audio application are increasing day-by-day on mobile devices. The continuity of these applications may hamper due to improper session rates during transmission. In this study we survey various papers for session rate prediction of streaming media using network traffic prediction methods. Also bandwidth estimation is carried out for the wireless network which plays a significant role in predicting the session rates. Our proposed session rate expression helps to understand the significance of predicting the session rate for streaming media in mobile wireless network. We study various proposals in this regard and the state-of-the-art analytical analysis is presented followed by our notations.