Classify With Caution: An Illustrative Example Using Mixture Models and Machine Learning

• In Event: Methods and Advancements in Mixture Modeling

Abstract

The study compared factor analytic techniques to machine learning classification algorithms using simulated data to assess various classification options for categorizing individuals into a relatively rare (10%) personality profile. The simulation manipulated factors such as the number of indicators, mean differences, variances, and item variances to evaluate ten classification techniques: traditional and Bayesian latent class analysis (LCA), 90/10 proportion, factor mixture models, classification trees, conditional inference trees, evolutionary trees, Ward's hierarchical clustering, Kmeans, and K-medians techniques. Overall, classification trees outperformed other methods in terms of Accuracy, Positive Predictive Value, and Sensitivity. However, none of the techniques achieved high classification accuracy for decision making. This study highlights the trade-off between explanatory and predictive utility and their limitations for individual decision making.

Authors

• Marcus A. Harris, University of Connecticut

• D. Betsy Mccoach, University of Connecticut