Classification of short-term flood events using stochastic variable selection and Gaussian Naïve Bayes classifier: A case study of Sirajganj district, Bangladesh.

Around the world, catastrophes caused by flooding are occurring naturally that cause a great deal of fatalities and financial loss. The loss of life and property can be considerably reduced with precise flood forecasts. The complexity of many flood predicting techniques makes the results difficult to interpret, compromising the process's core goal. This study uses a quick and flexible Gaussian Naïve Bayes (GNB) classifier to categorize eight different years as flooded or non-flooded based on predictor variables obtained via the Mutual Information (MI) technique. During the search, all-sky surface shortwave downward irradiance is identified as the optimum predictor variable out of nineteen stochastic variables, with the highest sensitivity for model accuracy. The model is then validated using four iterations derived from the MAPE of the GNB classification method for Twenty-five percent mean error rates from 4-fold cross-validation indicate that this classification model is suitable for flood forecasting. This high rate of mean error is caused by the short amount of data utilized as training data, as GNB requires huge data records to get effective results. This research could aid in the development and evaluation of hydrological projects in the Sirajganj district.