Machine learning assisted social system analysis: Youth transitions in five south and east Mediterranean countries.

While youth transitions to adulthood have been subject to various social studies, it is often the case that statistical tools of choice are limited in terms of sophistication and flexibility. Our study uses information collected as part of the SAHWA project [1] with the primary goal being to verify if machine learning can help rule out inappropriate assumptions and improve transition to adulthood analysis by outlining youth groups, their common characteristics and outlier cases (as well as if they are significant). As data includes numeric, as well as categorical and nominal variables use of common algorithms like K-Means clustering is not possible. It's also not reasonable to build on Eucledian distances in this mixed space, ruling out other classification methods that rely on it. We split the clustering algorithm selection into: (1) selection of distance calculation function, (2) algorithm and (3) decision on number of groups. A valuable information about transition to adulthood is obtained without imposing restrictive theoretical framework. [ABSTRACT FROM AUTHOR]