/amnesty_lm_lasso

Primary LanguageR

Amnesty International, Urgent Action (aiua) analysis

Testing linear/logistic regression models with regularization (lasso) for predicting Urgent Action follow-up

aiua_categories.R:

  • Summarizes urgent action category incidence with wordcloud and category incidence
  • i.e. shows overall frequencies of categories

aiua_logreg_fu_lasso.R

  • Given a category e.g. legal concern
    • Builds Corpus + DocumentTermMatrix for all UAs of given category
    • Runs logistic regression with 10-fold cross-validation to choose lambda (alpha=1 -> lasso)
    • Performance: mis-classification of whether an initial UA was followed up (<- requested outcome by AI)
    • e.g. Category 'Freedom of Expression': miss-classification : 0.21 on test data (all ~ 0.12 w/ stemming)
  • Output into figures/{category_name}