Weather Forecasting Modeling Using Soft-Clipping Swish Activation Function

Abstract

This research attempts to build on earlier work primarily based on activation functions, with a particular focus on network performance improvement using a new activation function called “Soft-Clipping Swish.” The goal of this research was to see how activation functions affected a weather forecasting model based on artificial neural networks (ANNs). The basic aim of this research, which underpins our approach, is to strengthen the experiments by diversifying them and extending them to timeseries from computer vision. When using that function, the negative side is completely ignored, while the right side retains the swish part of the function. This expansion was evaluated using a huge open-source dataset called Jena Climate, which is a weather timeseries dataset collected at the Max Planck Institute for Biogeochemistry's Weather Station in Jena, Germany.