Analysis of software development process in respect to anomaly detection

Abstract

The problem of detecting anomalous states or trends of changing the key design metrics is an actual problem in large software companies, which use software project management systems. The relevance of this task is conditioned by such factors: application of Agile software development methods; the amount of the project code; duration of the project (long-term projects); continuous improvement of the project functionality; stabilization of the system; improving the quality of the development process and the quality of the end product throughout all its life cycle. In this paper, a new method was proposed for searching for anomalous TS values through k-means clustering, using primary preprocessing - fuzzy transform (F-transform), which allows detecting the outliers not only in stationary TS but also in nonstationary TS extracted from software project management systems. This method is able to identify anomalies in the TS that characterized by strong oscillatory changes in the trend behavior or identify single atypical TS values. This method can be used for quickly localizing sections of TS trends atypical behavior for excluding such values in further analysis. In addition, this method can be applied iteratively, until the complete exclusion of values that clearly do not correspond to the behavior of TS tendencies (elementary, local, general).