Aim of the study: This research investigates the performance of Artificial Neural Network (ANN) and Ensemble models in predicting flood occurrence in the north-central states of Nigeria.
Methods: Meteorological data from the NASA POWER website and flood occurrence data from the Centre for Research on the Epidemiology of Disasters websites were collected. The collected data spanned from 1990 to 2022 for Benue, FCT-Abuja, Kogi, Kwara, Nassarawa, Niger, and Plateau states. The collected data are the input parameters in training the machine learning models: Artificial Neural Networks (ANN), Adaptive Boosting (AdaBoost), Stochastic Gradient Boosting (GBM), Extreme Gradient Boosting (XGBoost), and Random Forest (RF) models, for predicting flood occurrence in the region. The metrics used in evaluating the performance of the models were accuracy score, mean absolute error (MAE), and root mean squared error (RMSE).
Results: The five (5) models achieved high accuracy, with the GBM model demonstrating the best performance by achieving an accuracy of 87.14%, MAE of 0.13, and RMSE of 0.36. The findings also showed that flooding occurs when precipitation is above 500mm. Sensitivity analysis shows that precipitation and minimum temperature are the most significant parameters influencing the model predictions.
Conclusion: ANN and Ensemble models are efficient machine learning techniques for predicting flood occurrence in north-central Nigeria. The GBM model can yield reliable flood predictions for the region.
Key words: Flood prediction, Artificial Neural Network, Ensemble Model, XGBoost, Stochastic Gradient Boosting model, Random Forest, AdaBoost, Machine Learning, Deep Learning
|
