Use time0 data (pre-campaign) to predict time1 data (7-14 days ahead).
- model: gradient-boosted decision trees (light GBM)
- left panel: simple bivariate correlation between time0 (x) and time1 (y) values
- right panel: model-predicted time1 value (x) and observed time1 value (y)
# Predict mean (quality)
![[lgbm_quality_mean_y.png]]
# Predict sum (quality)
![[lgbm_quality_sum_y.png]]
# Predict count (no. of tweets)
![[lgbm_quality_count_y.png]]