| Abstrakt: |
Floods pose diverse challenges for organisms, infrastructure, and the environment, especially with the increasing frequency of natural disasters. Many existing works focused on traditional flood prediction methods, relying on statistical models which struggle in adapting to the dynamic environmental shifts, resulting in inaccurate forecasts and impeding effective disaster response. To address this, we propose a flood prediction paradigm that integrates both statistical models and social media data analysis techniques. By employing ensemble models combining classifiers such as K-Nearest Neighbor (KNN), Logistic Regression, Random Forest for statistical data analysis, alongside advanced natural language processing techniques such as Bidirectional Encoder Representations from Transformers (BERT), Long Short Term Memory (LSTM), and FastText for tweet analysis, our multimodal framework offers a comprehensive solution for flood forecasting. Leveraging rainfall data and user tweets, this study explores the fusion of diverse datasets to enhance predictive accuracy and provide timely insights into flood risk. The proposed system achieved an accuracy of 97%. Experimental results exhibit that the system outperforms other traditional methods. It captured the dynamic interplay between environmental data and human-generated content by amalgamating the predictive capabilities of diverse classifiers with advanced natural language processing techniques. |
Full text from SpringerLink