Software process line discovery

Abstract

Companies define software processes for planning and guiding projects. Since process definition is expensive, and in practice, no one process "fits all" projects, the current trend is to define a Software Process Line (SPrL): a base process that represents the common process elements, along with its potential variability. Specifying a SPrL is more expensive than just specifying one process, but the SPrL can be adapted to specific project contexts, minimizing the amount of extra work carried out by employees. Mining project logs has proven to be a promising approach for discovering the process that is applied in practice. However, considering all the possible variations that may be logged, the mined process may be overly complex. Some algorithms deal with this by filtering infrequent relations between log events, but they may discard relevant relations. In this paper we propose the v-algorithm that uses two thresholds to set up a SPrL: highly frequent relations are used to build the base process, variable relations define process variability, and rare relations are discarded as noise. We applied the $v$-$algorithm$ to the project log of Mobius, a small Chilean software company. We obtained a SPrL where we identified unexpected alternative ways of performing certain activities, as well as an optional activity that was originally specified as mandatory.