• Panel Data - The data given is usage metrics for different accounts/users at different time periods, this is called panel data in econometrics. Found some resources about using LSTMs for the problem but is not trivial, these problems also have to do with interpretability and interventions required. Many features for different accounts thus don't follow normality etc. Trying out a few approaches like imbalanced classification, unsupervised anomaly detection etc. Can interpret the results using SHAP etc.
  https://www.reddit.com/r/MachineLearning/comments/f9dhn5/d_any_cool_ml_longitudinal_data_analysis_methods/
  https://www.reddit.com/r/datascience/comments/erxsc4/what_is_the_contrary_of_time_series_data_called/