Talk · International Scientific-Technical Conference on Actual Problems of Electronic Instrument Engineering · November 14, 2025

Evaluation of Accuracy and Performance of Neural Network and Machine Learning Models for Flood Damage Prediction

Kuan Madiyarov, Пестунов Андрей Игоревич

Куан Мадияров

Open via DOI

Abstract

Flood damage prediction is essential for planning mitigation strategies and allocating resources. This study compares four regression models. The models considered in this study include Random Forest, Gradient Boosting, XGBoost, and a feedforward neural network – for predicting flood damage (in thousand USD) using a historical disaster dataset (1900–2025) based on EM-DAT records. We evaluate each model’s accuracy and performance (training time and memory usage). The dataset includes multifaceted features (hazard intensity, socio-economic indicators, etc.), and all models are trained and tested under a consistent pipeline with standardized preprocessing (including numeric feature scaling and one-hot encoding of categorical variables). Model quality is reported using the canonical regression criteria—Mean Absolute Error, Root Mean Squared Error, and the coefficient of determination. Across experiments, tree-ensemble approaches, with gradient boosting in particular, delivered the strongest predictive accuracy, whereas the feedforward neural network was less competitive on this tabular dataset. Gradient Boosting slightly outperforms Random Forest in coefficient of determination (0.74 vs 0.73) with the lowest RMSE, and also trains an order of magnitude faster than the neural network. The XGBoost model also exhibits strong performance, comparable to Random Forest. We discuss the implications of these findings in the context of existing literature and outline how model choice can impact flood risk assessment.

Flood Risk Assessment and ManagementHydrological Forecasting Using AIEnergy and Environmental Systemsflood damage predictionflood risk assessmentEM-DAT

Evaluation of Accuracy and Performance of Neural Network and Machine Learning Models for Flood Damage Prediction

Abstract

Related publications