Design and Development of Automated Penetration Testing Model for Network Efficacy
| dc.contributor.guide | Joshi, Shubhalaxmi | |
| dc.coverage.spatial | ||
| dc.creator.researcher | Railkar, Dipali Nilesh | |
| dc.date.accessioned | 2025-12-15T04:08:12Z | |
| dc.date.available | 2025-12-15T04:08:12Z | |
| dc.date.awarded | 2025 | |
| dc.date.completed | 2025 | |
| dc.date.registered | 2021 | |
| dc.description.abstract | The increasing complexity of cyber threats necessitates intelligent and scalable penetration testing frameworks that go beyond manual efforts. Traditional approaches are resource-heavy and often limited by rigid configurations and lack of adaptability. This research introduces a two-stage development of automated penetration testing models integrating reinforcement learning (RL) and deep learning (DL) to enhance detection accuracy and testing efficiency. newlineInitially, a PNS-optimized Q-learning Ensemble Deep CNN (PNS-QCNN) was developed, which combines Convolutional Neural Networks with a reinforcement learning controller and Prairie Natural Swarm optimization for fine-tuning parameters. This model significantly improves classification accuracy and performance in simulated network environments. Building upon this, the Incremental Q-Learning Ensemble Deep CNN (IQLEDN) was proposed to further address the gaps of static learning and prioritization. IQLEDN incorporates fuzzy logic for semantic vulnerability assessment, allowing classification into risk-based categories (None, Partial, Complete) using CVSS and exploitability metrics. It also integrates Canidae-Procyonid Swarm (CPrS) optimization and adopts an incremental learning mechanism to adapt over successive training iterations. newlineBoth models were trained on a structured CVE dataset comprising real-world vulnerabilities. The IQLEDN model achieved superior results with an accuracy of 95.90% on 90% training data and 95.96% with 10-fold cross-validation, outperforming state-of-the-art classifiers like KNN, CatBoost, XGBoost, and LSTM. Performance metrics and cumulative reward analysis confirmed the learning efficiency and adaptability of the models. newlineThis research contributes a scalable, accurate, and intelligent framework for penetration testing, filling key gaps in existing systems. Future work will explore real-time deployment, richer datasets, and further explainability to support critical cybersecurity applications. newline | |
| dc.description.note | ||
| dc.format.accompanyingmaterial | None | |
| dc.format.dimensions | ||
| dc.format.extent | ||
| dc.identifier.researcherid | ||
| dc.identifier.uri | http://hdl.handle.net/10603/680643 | |
| dc.language | English | |
| dc.publisher.institution | School of Computer Engineering and Technology | |
| dc.publisher.place | Pune | |
| dc.publisher.university | Dr. Vishwanath Karad MIT World Peace University | |
| dc.relation | ||
| dc.rights | university | |
| dc.source.university | University | |
| dc.subject.keyword | Computer Science | |
| dc.subject.keyword | Computer Science Software Engineering | |
| dc.subject.keyword | Engineering and Technology | |
| dc.title | Design and Development of Automated Penetration Testing Model for Network Efficacy | |
| dc.title.alternative | ||
| dc.type.degree | Ph.D. |
Files
Original bundle
1 - 5 of 12
Loading...
- Name:
- 01_title.pdf
- Size:
- 32.21 KB
- Format:
- Adobe Portable Document Format
- Description:
- Attached File
Loading...
- Name:
- 02_prelim pages.pdf
- Size:
- 501.14 KB
- Format:
- Adobe Portable Document Format
License bundle
1 - 1 of 1