FLOOD FORECASTING IN MALAYA ALMATINKA RIVER VIA MACHINE LEARNING AND DEEP LEARNING WITH OVERSAMPLING

Abstract

Flooding, a phenomenon characterized by the overflow of water from its natural confines onto dry land, poses significant threats to communities and infrastructure, often resulting from heavy precipitation, snow melting, and various natural and anthropogenic factors. The causes of flooding encompass a myriad of influences, including intense rainfall, precipitation patterns, and meltwater accumulation. Such events precipitate abrupt rises in river and lake levels, accompanied by the formation of barriers. The breaching of dams and levees can trigger the rapid propagation of large volumes of water, generating formidable breach waves.

In contemporary flood management practices, machine learning and deep learning algorithms have emerged as indispensable tools for forecasting and mitigating flood risks. This study focuses on predicting floods in the Malaya Almatinka River, situated in Almaty, Kazakhstan. Leveraging a diverse set of algorithms including XGBoost, LightGBM, RandomForest, SVM, Linear Regression, and neural networks, the research endeavors to enhance flood prediction accuracy. However, during the data preprocessing phase, it was observed that the dataset suffered from imbalance, necessitating the implementation of Random Over-Sampling to rectify the issue and ensure more equitable representation across classes. Through the fusion of advanced computational techniques and empirical data, this research aims to contribute towards more effective flood forecasting strategies, thereby bolstering the resilience of communities in flood-prone regions.