Analytical Software Engineering: A Novel Design Paradigm for Problem Formulation and Optimization

Software engineering faces increasing challenges due to the growing complexity of modern systems, requiring innovative solutions for design optimization, maintainability, and performance enhancement. Traditional approaches often fail to comprehensively address these challenges, particularly in design pattern detection and search-based refactoring. This thesis introduces and examines the design paradigm of Analytical Software Engineering (ASE) as a pioneering framework to bridge this gap. By emphasizing abstraction, compatibility, and scalability, ASE provides a pioneer methodology for modeling and solving complex software engineering problems, enabling innovative solutions for critical tasks. ASE's core contributions include Behavioral-Structural Sequences (BSS) and the Optimized Design Refactoring (ODR) frameworks. The BSS encapsulate software artifacts into a compact, languageagnostic format, enabling precise design pattern detection using machine learning models such as transformers, this abstraction preserves essential structural and behavioral details while ensuring compatibility with diverse computational tools. The ODR framework unifies artifact and solution representations, streamlining heuristic-driven refactoring and eliminating computational overhead from traditional iterative approaches. Empirical validation demonstrates ASE's feasibility, effectiveness, and generality as design pattern detection in code achieves high accuracy through transformer-based models trained on BSS, while the ODR framework demonstrates efficient optimization through heuristic algorithms, such as genetic algorithms. ASE effectively balances granularity and abstraction, retaining critical problem details while abstracting extraneous complexity. Limitations, such as challenges in encoding certain metrics like encapsulation or line-of-code attributes, point to opportunities for future research.