Uncertainty-wise software anti-patterns detection: A possibilistic evolutionary machine learning approach

Code smells (a.k.a. anti-patterns) are manifestations of poor design solutions that can deteriorate software maintainability and evolution. Existing works did not take into account the issue uncertain class labels, which is an important inherent characteristic detection problem. More precisely, two human experts may have different degrees uncertainty about smelliness a particular only for smell task but also type identification one. Unluckily, existing approaches usually reject and/or ignore data correspond to classes (i.e. dataset instances) with labels. Throwing away disregarding factor could considerably degrade detection/identification process effectiveness. From solution approach viewpoint, there no work in literature proposed method able detect identify code while preserving aspect. The main goal our research handle factor, issued from experts, detecting identifying by proposing evolutionary deal anti-patterns classification We suggest Bi-ADIPOK, as effective search-based tool capable tackle previously mentioned challenge both cases. corresponds EA (Evolutionary Algorithm) optimizes set detectors encoded PK-NNs (Possibilistic K-nearest neighbors) based on bi-level hierarchy, upper level role consists finding optimal parameters, lower one generate PK-NNs. A newly fitness function has been PomAURPC-OVA_dist modified Area Under Recall Precision Curve One-Versus-All_distance, abbreviated PAURPC_d this paper). Bi-ADIPOK label using some concepts stemming Possibility Theory. Furthermore, even imbalanced data. notice first built then validated possibilistic base examples simulates mimics subjectivity engineers opinions. statistical analysis obtained results comparative experiments respect four relevant state-of-the-art methods shows merits proposal. demonstrate that, environment, ranges between 0.902 0.932 its IAC lies 0.9108 0.9407, certain 0.928 0.955 0.9477 0.9622. Similarly, results, indicate varies 0.8576 0.9273 0.8693 0.9318. For 0.8613 0.9351 values 0.8672 0.9476. With data, find 35% more than second best (i.e., BLOP). succeeded reduce number false alarms misclassified smelly 12%. In addition, 43% types BLOP reduces 32%. same demonstrating Bi-ADIPOK’s ability such environment.