Weather Parameters and Heat Index Prediction Applying Deep Neural Networks

Abstract

The way through which change of weather parameters is measured can be called as weather forecasting. Weather forecast plays important role in predicting natural calamities, agricultural sectors, some industrial sectors, etc. Heat index is a vital part of weather which depends on some weather parameters. Human body temperature, evaporation, etc. depends on heat index. Previously, several researchers worked on this issue to determine whether it rains or not for two to three days. Some of them worked just to determine the effect of change of heat index in different regions. A number of works were based on machine learning algorithms and most of them had either low accuracy or high error rate. But in this paper, we proposed a combined deep learning model. It can predict five most important weather parameters up to 30 days with very low error rate by using single cell long short term memory (LSTM). Along with this, the output from LSTM was further used to predict heat index using artificial neural network (ANN) which also gave a very high accuracy. The dataset provided to the model was pre-processed properly by using Gaussian filter, Median filter and scaling. It increased the sensitivity and performance of the model. We got the mean absolute percentage error rate (MAPE) ranging from 0.02%-8.53% with LSTM model and 94.68% of accuracy from ANN model. Further judging parameters of ANN are precision (94.57%), recall (94.57%), f1 score (94.57%), loss (12.75%). The other error rates of LSTM model are MSE (1%-16%), RMSE (10.72%-39.44%), R2 (99%-100%), MAE (8.17%-34.59%).