Evolution of Long Short-Term Memory (LSTM) in Air Pollution Forecasting

Abstract

Long short-term memory (LSTM) networks are a tweaked version of the recurrent neural networks (RNN) that enable information persistence and fall under the deep learning domain. It’s widely used in various applications, such as air pollution forecasting, flood forecasting, handwriting generation, language modeling, image captioning, question answering, video to text conversion, machine translation, etc. LSTM networks are used to process sequential data that involves the temporal correlation between a given data segment and its previous segment. The chapter starts with a discussion of the LSTM architecture, its variants, and its applications across various domains. The chapter also provides a comprehensive discussion of various types of LSTMs being used to solve the problem of air pollution forecasting, along with discussing a general pipeline used for the same.