Tejas Shah's research while affiliated with L. D. College of Engineering and other places

Weather forecasting is fraught with peril. Accurate predictions are now possible thanks to the (NWP) models' rapid development over the past few decades. However, the surroundings are absolutely disorganized. Many people, therefore, reject making use of value estimations unless an associated uncertainty estimate is also offered. Currently, the only option to generate a confidence estimate for individual forecasts is to create an ensemble of numerical weather simulations, which is computationally pretty expensive. This study explores the possibility that machine learning approaches can provide an alternate approach to estimating the degree to which future weather predictions may deviate from the baseline. Preprocessing, feature selection, and model training make up the first three steps of the suggested methodology. Data preprocessing includes all cleaning, integration, selection, transformation, and reduction processes. To choose features, they employ autocorrelation, partial autocorrelation, and PCA. Following feature selection, the models are subsequently trained using CNN-GRU. The two most popular alternatives, GRU and CNN, are outperformed by the suggested strategy.