Big Data Analytics using Deep LSTM Networks: A Case Study for Weather Prediction

Shweta Mittal, Om Prakash Sangwan

Advances in Science Technology and Engineering Systems Journal · 01 January 2020

In one sentence. Deep LSTM networks trained on 15 years of hourly Brazilian station data for 1-hour and 24-hour weather prediction; two hidden layers balance accuracy and runtime.

Abstract

Recurrent Neural Networks has been widely used by researchers in the domain of weather prediction.Weather Prediction is forecasting the atmosphere for the future.In this proposed paper, Deep LSTM networks has been implemented which is the variant of RNNs having additional memory block and gates making them capable of remembering long term dependencies.Fifteen years hourly meteorological data of Brazil weather stations for the period of 2000-2016 collected from Kaggle.com has been analyzed for 1 hour and 24-hour time lag using Keras libraries on Spark framework.Hidden layers of the network have been increased up to three to examine its impact on accuracy of the network and it was found that network with 2 hidden layers provides good accuracy in lesser learning runtime.From the experimental results, it is also concluded that Adam optimizer provides best results when compared with SGD and RMSProp optimizer.

Labels

Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.