In the last few years, several new tools addressing maturity level management have been proposed, e.g. diagnostic assessment questionnaires (DAQ). In practice, the usage of questionnaires presents some drawbacks related to subjectivity, time cost, and applicant bias. Moreover, the questionnaires may present a large number of questions, as well as part of them redundant. Another important fact of real-life application of DAQs concerns the usage of multiple questionnaires, increasing the shortcoming impacts. To pave the way to a more convenient tool to support and facilitate the achievement of organizational strategies and objectives, we proposed an intelligent reduction of DAQs by the use of single-label and multilabel feature selection. In this paper, we reduced four DAQs (Risk Management, Infrastructure, Governance and Service Catalogs) with our proposal in comparison to different feature selection algorithms (χ2, Information Gain, Random Forest Importance and ReliefF). The reduction was driven by a machine learning prediction model towards ensuring the new subset of question grounded in the same obtained score result. Results showed that removing irrelevant and/or redundant question it was possible to increase the model fitting even reducing about one-third of the questions with the same predictive capacity.
