Deep Learning for Feature Engineering-Based Improved Weather Prediction

Abstract

Weather prediction has gained a point of attraction for many researchers of variant research communities. The emerging deep learning techniques have motivated many researches to explore hidden hierarchical pattern in the great mass of weather dataset for weather prediction. In this chapter, four different categories of computationally efficient deep learning models—CNN, LSTM, CNN-LSTM, and ConvLSTM—have been critically examined for improved weather prediction. Here, emphasis has been given on supervised learning techniques for model development by considering the importance of feature engineering. Feature engineering plays a vital role in reducing dimension, decreasing model complexity as well as handling the noise and corrupted data. Using daily maximum temperature, this chapter investigates the performance of different deep learning models for improved predictions. The results obtained from different experiments conducted ensures that the feature engineering based deep learning study for the purpose of predictive modeling using time series data is really an encouraging approach.