River flooding is a significant natural hazard that can cause substantial damage to society. This paper presents a novel approach for a multi-cluster-based river flood early warning system (EWS) using fuzzy-logic-based learning (FLBL) and fuzzy rule optimization. The proposed system consists of cluster-based IoT nodes that collect sensor data from water level and rainfall sensors, and a server-side component that handles missing data imputation, FLBL for generating fuzzy rules, and an inference engine for predicting incoming flood events. The FLBL technique automatically learns and generates optimized fuzzy rules from the dataset, eliminating the need for manual trial-and-error processes. The system was showcased in three districts in Semarang City, Indonesia, and the results demonstrated that the proposed method outperformed baseline methods using decision trees and neural networks, achieving an average accuracy of 97.87% in predicting flood events. The proposed approach addresses the challenges of multi-cluster-based EWS and provides a reliable solution for early flood event prediction.