Flooding is one of the most devastating natural disasters globally, causing extensive damage to infrastructure, the environment, and human lives. With increasing occurrences due to climate change, accurate classification and analysis of flood imagery are essential for early detection, damage assessment, and post-disaster recovery. Reliable flood classification systems are critical for early warning, resource allocation, and mitigation efforts, helping to minimize the impact on affected regions. Remote sensing and computer vision techniques, including the Bag-of-Visual-Words (BOV) model, offer powerful tools for interpreting flood images by categorizing and identifying flooded regions across vast and complex terrains. This paper presents a modification of the standard Random Forest algorithm to enhance the accuracy of image classification within a Bag-of-Visual-Words (BOV) model. The modified Random Forest achieves better adaptability and performance across flood image datasets by introducing flexibility in parameter tuning through custom hyperparameters and automatic grid search. This modification addresses challenges in balancing efficiency and accuracy for classifying high-dimensional image data sets.