Table 1 Summary of reviewed studies.
Authors | Applications | Traditional method | Deep learning |
|---|---|---|---|
Salama and El-Gohary17 | Rule extractions | SVM: 70 NB: 44 ME: 82 | - |
Goh and Ubeynarayana18 | Construction accident types | SVM: 67 LR: 35 RF: 42 KNN: 44 DT: 53 NB: 41 | - |
Zhang et al.19 | Construction accident types | DT: 52 KNN: 53 NB: 44 SVM: 58 LR: 50 Ensemble: 52 Optimized ensemble: 68 | - |
Ul Hassan et al.20 | Design-build contract requirements | SVM: 93.17 NB: 87.48 LR: 94.18 DT: 89.38 k-NN: 92.03 ANN: 92.80 | - |
Luo et al.21 | Construction accident types | SVM: 73 NB: 54 LR 66 | CNN: 76 |
Wang et al.22 | Construction defects categories | DT: 92% RF: 91% NB: 98% SVM: 85% | CNN-AT: 92% |
Fang et al.23 | Safety requirements on construction sites | - | TextCNN with BiGRU: 75.31 TextCNN: 74.82 BiGRU with Attention: 75.18 TextRCNN: 75.26 BERT: 86.91 |
Yang et al.24 | Facility defects in residential buildings | - | CNN (FastText): 87.44 CNN (Word2Vec): 88.25 BERT: 89.62 ELECTRA: 89.47 GPT2: 89.68 CNN (FastText and Word2Vec): 90.72 |
Wang et al.25 | Safety knowledge mapping | BERT-BiLSTM: 75.48 BERT-LSTM-CRF: 91.08 BERT-BiLSTM-CRF: 91.74 | |
Tian et al.26 | Safety hazard classification | TextCNN: 79.56 BiLSTM: 80.42 BERT-GCN- BiLSTM: 86.56 BERT-GCN: 82.34 | |
Jianan et al.27 | Knowledge types of consulting standards | BERT: 90.17 RoBERTa: 90.94 Longformer-RoBERT: 91.65 Longformer-RoBERT with CNN: 86.7 Longformer-RoBERT with LSTM: 83.81 | |
Construction contract disputes | SVM: 62.65 NB: 50.55 LR: 62.50 DT: 57.90 k-NN: 44.85 | TextCNN: 65.99 TextCNN + LSTM: 63.20 R-CNN: 57.91 Transformer: 54.80 | |
Aim of this research | Fire door defect classification | ANN, SVM, DT, LR, 1D CNN, and LSTM | BERT, RoBERTa, ALBERT, DistilBERT, and XLNet |
