Artificial Intelligence Models Extracting Different Structures From The Same Data Set

Psychometric surveys are rich data source used to measure human behavioral properties. But artificial intelligence and machine learning researchers face a new question: Are different representation learning methods reach the same results from the same data?

A research team reviewed the data obtained from 757 teachers-child pairs to answer this question. The study applied different analysis techniques on child behavior scales (SDQ, ASBI, CBCL):

• The basic Component Analysis (PCA) and the size reduction method produced different results
• Clustering algorithms reveal behavioral grouping on the same data set
• Information at Article level is interpreted differently according to the selected mathematical approach
• The research provided statistical meaningfulness with 757 dual data

These findings show that artificial intelligence models are dependent on the selected method rather than in the process of data analysis. The reliability of psychometric research can be determined not only by the quality of the data, but also with the selection of the analysis method.

As a result, researchers who measure behavioral properties should be aware of how they shape the results of the machine learning technique they use.