Determining Disaster Risk Management Priorities Through a Neural Network-Based Text Classifier

Community participation and involvement plays a big role in disaster risk reduction. This paper made use of the feedback from public on how local communities can be better prepared in times of disaster. Main goal of this study is to automatically assign qualitative responses into its appropriate category in disaster management using bidirectional recurrent neural network. In building the BRNN model, data corpus was split into training set (85%) and testing set (15%), which achieved acceptable average accuracy rate of 81.67%, 81.17% precision, 81.67% recall and 80.81% f-measure. Output of the classifier showed that the top four priority needs of the respondents in DRR fall under the categories of education and training; communication and coordination; dissemination of information alerts and warnings/early warning system; and role of local authority. The validated results generated is a useful feedback to concerned agencies, specifically in the Province of Albay in enhancing their existing disaster management plans. Future work may add trained data to achieve higher performance results. Using other hyperparameters in the configuration of neural networks may also be considered for better evaluation result of the classification model.