A BERTweet-based design for monitoring behaviour change based on five doors theory on coral bleaching campaign

Journal of Big Data

Table 7 BERTweet stacking ensemble (SE) result

Scenario	Description	Accuracy	F1 score
SE#1	modelSE#1	0.7429	0.6786
SE#2	modelSE#2	0.7592	0.7399
SE#3	modelSE#3	0.7673	0.7473
SE#4	modelSE#4	0.7429	0.7436
SE#5	combination1 (modelSE#1 and modelSE#2)	0.7755	0.7677
SE#6	combination2 (modelSE#1 and modelSE#3)	0.7551	0.7530
SE#7	combination3 (modelSE#1 and modelSE#4)	0.7551	0.7298
SE#8	combination4 (modelSE#2 and modelSE#3)	0.7388	0.7236
SE#9	combination5 (modelSE#2 and modelSE#4)	0.7633	0.7829
SE#10	combination6 (modelSE#3 and modelSE#4)	0.7592	0.7754
SE#11	combinationAll (modelSE#1—modelSE#4)	0.7796	0.7945

Scenario SE#3 is bolded because it has the best performance out of all standalone models, scenario SE#9 is bolded because it has the best performance out of the combination of two models and SE#11 is bolded because it has the best performance out of the combination of four models