Empirical validation of structural metrics for predicting understandability of conceptual schemas for data warehouse

Abstract

Data warehouse (DW) quality depends on its data models (conceptual, logical and physical model). Multidimensional (MD) modeling has been widely recognized as the backbone of data modeling for DW. Recently, some of the authors have proposed a set of structural metrics to assess quality of MD conceptual models. They have found the significant relationship between metrics and understandability of DW conceptual schemas using various correlation analysis techniques such as Spearman’s, Pearson etc. However, advanced statistical and machine learning methods have not been used to predict effect of each metric on understandability. In this paper, our focus is on predicting the effect of structural metrics on understandability of conceptual schemas using (i) statistical method (logistic regression analysis) that include univariate and multivariate analysis, (ii) machine learning methods (Decision Trees, Naive Bayesian Classifier) and (iii) compare the performance of these statistical and machine learning methods. The results obtained show that some of the metrics individually have a significant effect on the understandability of MD conceptual schema. Further, few of the metrics have a significant combined effect on understandability of conceptual schema. The results also show that the performance of Naive Bayesian Classifier prediction method is better than logistic regression analysis and Decision Trees methods.