Dataset for estimating overtime allocation in Sodtware Development Projects

Software overtime has been studied using multi-objective optimization techniques without paying attention to project managers' (PMs) preferences.  This dataset, comprising 1,622 instances of overtime planning solutions generated from six software projects of varying sizes and annotated by 20 software PMs, is produced to build machine learning models for estimating project managers' satisfaction (evaluation score) with overtime plans for software projects. The dataset is constructed from six software projects earlier collected by DeO Barros and De Araujo (2016), and obtained from https:\/\/github.com\/luizaraujojr\/GECCO2016.  Work package data were extracted from the project descriptions (originally in XML) to build schedules for the projects. Then, the search-based multi-objective approach was used to generate optimal overtime plan solutions, which were passed to PMs for annotation (a score between 1 and 100 denoting satisfaction level). The annotations collected from PMs were further analyzed to remove outliers. The final score is the mean of the remaining annotations. The dataset can be utilized to train ML models for use in real-world overtime estimation for similar projects. It is also useful for building  interactive optimization algorithms in overtime planning.