Proposed extended analytic hierarchical process for selecting data science methodologies

Abstract

Decision making can present a considerable amount of complexity in competitive environments; where methods that support possess great relevance. The article presents an extension of the Hierarchic Analytical Process; complemented with Personal Construct Theory, which purpose is to reduce ambiguity when defining and establishing values for the criteria in a determined problem. In recent years, the scope for decision making based on data has considerably raised, which is why Data Science as a scientific field is rising in popularity; where one of the main activities for data scientists is selecting an adequate methodology to guide a project with this traits. The steps defined in the proposed model guide this task, from establishing and prioritizing criteria based on degrees of compliance, grouping them by levels, completing the hierarchical structure of the problem, performing the correct comparisons through different levels in an ascendant manner, to finally obtaining the definitive priorities of each methodology for each validation case and sorting them by their adequacy percentages. Both disparate cases, one referred to an industrial/commercial field and the other to an academic field, were effective to corroborate the extent of usefulness of the proposed model; for which in both cases MoProPEI obtained the best results.