Table 3 Evaluation of classification model on Jakarta Traffic Tweet Corpus
From: Traffic and road conditions monitoring system using extracted information from Twitter
NB | SVM | LR | RF | |
|---|---|---|---|---|
Relevance prediction | ||||
Count vector, unigram | 87.44% | 89.20% | 90.55% | 88.88% |
TF-IDF, unigram | 86.54% | 90.00% | 89.73% | 88.96% |
TF-IDF, bigram | 79.32% | 88.34% | 88.17% | 87.45% |
TF-IDF, trigram | 62.63% | 82.52% | 82.73% | 82.21% |
TF-IDF, char {1,2} gram | 82.46% | 90.40% | 90.00% | 86.88% |
TF-IDF, bigram + trigram | 85.81% | 90.23% | 90.04% | 89.30% |
Unigram + char gram | 87.35% | 89.65% | 90.72% | 87.49% |
Traffic event type prediction | ||||
Count vector, unigram | 82.07% | 93.79% | 94.04% | 90.10% |
TF-IDF, unigram | 81.51% | 93.93% | 92.35% | 89.93% |
TF-IDF, bigram | 84.72% | 89.60% | 89.10% | 87.64% |
TF-IDF, trigram | 77.46% | 79.17% | 79.26% | 77.80% |
TF-IDF, char {1,2}gram | 76.30% | 92.66% | 91.41% | 90.06% |
TF-IDF, bigram + trigram | 83.51% | 93.93% | 92.68% | 91.81% |
Unigram + char gram | 81.72% | 93.85% | 94.14% | 89.99% |