Unsupervised Learning of the Sequences of Adulthood Transition Trajectories

Abstract

Extracting patterns from a complex real-life dataset and drawing inferences, thereafter, are becoming part and parcel in various areas of research during the past two decades. Unsupervised machine learning is a type of self-organized learning, which helps to find previously unknown patterns in the real-life dataset without pre-existing labels. However, analysing, understating and identifying the typical sequential patterns of events in complex event history data and the ability to utilize this retrieved knowledge create a significant impact on many aspects of individual life course. This paper first introduces the sequence analysis for analysing the adulthood and family formation sequences to better understand the evolution, features and typologies of various complex life course processes. Not only the ordering of sequences, grouping these existing sequential patterns into clusters is also challenging in life course analysis. Modern tools of unsupervised machine learning are an appropriate choice to analyse the sequences of important life course trajectories and in particular, when there is unobserved heterogeneity present in the data. The present article uses life course retrospective data of Indian youths that give adulthood transition trajectories from six Indian states. We estimate and interpret the similarity and distances between sequences using the optimal matching approach. Cluster analysis has been used in order to produce prominent typologies of the adulthood event sequence trajectories. To conclude, unsupervised learning of the sampled sequences of adulthood transitions considered herein has successfully demonstrated its potential usefulness in displaying and summarizing complex event history data into meaningful and interpretable dimensions to meet new challenges and to build policy framework for the adults of a nation.