This paper presents a flood inundation prediction method based on dimensionality reduction and deep learning, and evaluates its performance through comparison with observed inundation data. Numerical flood simulations are first conducted, and the resulting inundation depth distributions are transformed into low-dimensional coefficients using a dimensionality reduction technique. A deep learning model is then trained to predict these coefficients from overtopping and levee breach water depths as input conditions, and inundation area and water depth distributions are reconstructed from the predicted coefficients. This framework enables rapid estimation of inundation distributions within a few seconds with high accuracy. The proposed method is applied to the Kinu River basin, which experienced severe flooding during the Kanto–Tohoku heavy rainfall event in September 2015. A large number of inundation depth datasets are generated through numerical flood simulations and used to construct a training dataset. The predicted inundation area and water depth are compared with observed inundation data, and the performance of the method is evaluated based on the agreement of inundation area and the error in inundation depth. The results show that the proposed method can reasonably reproduce the observed inundation characteristics, demonstrating its potential applicability to real-time flood forecasting and emergency response support.