Using Machine Learning Algorithm to Analyze Longitudinal Data: Association between Job Satisfaction and Job-Related Variable in the UK Household Longitudinal Study(UKHLS)

In this study we follow the tutor guide of Steetal et al (2022) to use machine learning algorithms
to investigate the important features associated with job satisfactions and the prediction power
with data from the UK longitudinal household study (UKHLS). Using the job historical data span
from 2010 to 2023 in UKHLS we compared several machine learning algorithms, such as random
forest, support vector regression and XGBoost with traditional methods such as linear regression or
Bayesian regression. The machine learning algorithms in general have better prediction power than
linear model or Bayesian regression, but the difference is limited. More important, the framework
suggested by Steetal has distorted the temporary structure of longitudinal data and therefore the
prediction cannot be explained in real term. More suitable machine learning methods should be
proposed for longitudinal data analysis.