Date of Award

8-2022

Document Type

Thesis

Degree Name

Master of Science (MS)

College/School

College of Science and Mathematics

Department/Program

Computer Science

Thesis Sponsor/Dissertation Chair/Project Chair

Vaibhav Anu

Committee Member

Aparna Varde

Committee Member

Kazi Zakia Sultana

Abstract

Gathering and extracting security requirements adequately requires extensive effort, experience, and time, as large amounts of data need to be analyzed. While many manual and academic approaches have been developed to tackle the discipline of Security Requirements Engineering (SRE), a need still exists for automating the SRE process. This need stems mainly from the difficult, error-prone, and time-consuming nature of traditional and manual frameworks. Machine learning techniques have been widely used to facilitate and automate the extraction of useful information from software requirements documents and artifacts. Such approaches can be utilized to yield beneficial results in automating the process of extracting and eliciting security requirements. However, the extraction of security requirements alone leaves software engineers with yet another tedious task of prioritizing the most critical security requirements. The competitive and fast-paced nature of software development, in addition to resource constraints make the process of security requirements prioritization crucial for software engineers to make educated decisions in risk-analysis and trade-off analysis.

To that end, this thesis presents an automated framework/pipeline for extracting and prioritizing security requirements. The proposed framework, called the Security Requirements Extraction and Prioritization Framework (SecREP) consists of two parts:

  • SecREP Part 1: Proposes a machine learning approach for identifying/extracting security requirements from natural language software requirements artifacts (e.g., the Software Requirement Specification document, known as the SRS documents)
  • SecREP Part 2: Proposes a scheme for prioritizing the security requirements identified in the previous step.

For the first part of the SecREP framework, three machine learning models (SVM, Naive Bayes, and Random Forest) were trained using an enhanced dataset the “SecREP Dataset” that was created as a result of this work. Each model was validated using resampling (80% of for training and 20% for validation) and 5-folds cross validation techniques. For the second part of the SecREP framework, a prioritization scheme was established with the aid of NLP techniques. The proposed prioritization scheme analyzes each security requirement using Part-of-speech (POS) and Named Entity Recognition methods to extract assets, security attributes, and threats from the security requirement. Additionally, using a text similarity method, each security requirement is compared to a super-sentence that was defined based on the STRIDE threat model. This prioritization scheme was applied to the extracted list of security requirements obtained from the case study in part one, and the priority score for each requirement was calculated and showcased

File Format

PDF

Share

COinS