| Popis: |
Floods account for substantial economic losses to a country by causing damage to agriculture, households, livelihoods, infrastructures and so on. Development of an accurate predictive model with locally observable factors for indicating early warning of a monsoon river flood could help to minimize losses. Such a model should be readily available and accessible by people and local agencies. In this study, six well-known machine learning algorithms and regression models, including Decision Tree, Naive Bayes, Support Vector Machine (SVM), Artificial Neural Networks (ANN), Logistic Regression, and LASSO have been compared in terms of accuracy of flood prediction. Hydro-meteorological observed data of associated station of Manikganj in Bangladesh, have been fed into this model to learn how to predict monsoon flooding as instructed by the algorithm on which the respective model has been built upon. Each model has been trained with one set of water level time series data of associated station, but models for all the stations share the same set of spatio-temporal rainfall data, observed from a common station. ANN and SVM have a reputation for delivering good performance for many applications and LASSO was found to be appropriate according to a previous study. However, the results of this study indicate that the model based on Naive Bayes is best suited for monsoon river flood prediction. |
