Text Preprocessing Approaches in CNN for Disaster Reports Dataset

This study aims to compare the performance of the text-preprocessing methods namely automatic and semi-automatic preprocessing techniques in the CNN algorithm to carry out learning on disaster report dataset. The experimental results on the disaster dataset with a total of 200 records with the automatic text preprocessing technique produce an average accuracy of 0.81 and 1 with training data of 80:20 and 90:10. While in the optimize model that is semi-automatic text preprocessing approach (which is the author's proposed approach), the average accuracy obtained are 0.95 and 1 for dataset 80:20 and 90:10. The experimental results conclude that cleaning the dataset with the semi-automatic text preprocessing model can improve accuracy compared to the previous model. The proposed model will get convergence with 80:20 training data at epoch 20, batch size 5 and random state 34, while for dataset 90:10 the best convergence value at epoch 20–30.