Weather Forecast System using FeedForward Backpropagation Technique

Weather forecasting is an important task in disaster management, especially for a city like Chennai, which is commonly affected by cyclones, heavy rainfall, and heat waves. The proposed study investigates the enhancement of the accuracy and reliability of weather prediction models with the help of the introduction of FeedForward Neural Networks (FFNNs) with the use of Rectified Linear Unit (ReLU) activation functions. Statement: Traditional activation functions like the sigmoid function are not effective activating functions as they suffer from vanishing gradient problems that do not allow deep networks to perform. In order to avoid these complications, FFNNs with ReLU, which allows for high efficiency and sparsity, are used to process large-scale meteorological datasets. The model is trained on historical weather data that is collected and preprocessed, and its performance is evaluated using metrics such as Mean Squared Error (MSE) and Root Mean Squared Error (RMSE). The findings reveal that the proposed approach leads to a notable enhancement in short- term prediction skill, particularly regarding the key physical parameters of temperature and wind speed. The early stopping and dropout layers actually reduce overfitting. These results demonstrate the potential of FFNNs for transforming weather forecasting systems to provide actionable information on extreme weather events risk for disaster management and decision- making in regions exposed to extreme weather events. Our research builds on an expanding body of literature around optimizing neural networks for meteorological applications and suggests areas for further research that could improve robustness and scalability. Citation: Palak Bansal Institute of Higher Education Research, Mandi 174323, India Abstract Accurately predicting the weather remains challenging, leading to injuries and deaths across the globe due to natural disasters.