AN ENSEMBLE LEARNING FRAMEWORK FOR VISIBILITY PREDICTION IN INDO-GANGETIC REGION

Arkapal Panda, Tanmay Basu, Vaibhav Kumar

Keywords: Machine Learning, Climate, Visibility, Indo-Gangetic region

TL;DR: Machine Learning for Visibility Prediction in Indo-Gangetic Region

Abstract:

Visibility of an area affects all forms of transportation such as sea, surface and aviation which can further affect the economy of that
area. Thus it is very important to accurately estimate the visibility of an area for the upcoming days based on different parameters of
the past meteorological data, so that we can take precautions in case of poor visibility. Several machine learning techniques have been
already applied on different kinds of data sets to estimate the visibility. However, most of these methods could not perform reasonably
well and none of them were applied on the meteorological data of the Indo-Gangetic plane in India, which witnesses widespread
fog primarily during winter that badly impacts visibility and therefore transportation. In this spirit, a Extreme Gradient Boosting
(XGboost) based regression framework is developed here to estimate the visibility of the Indo-Gangetic plane. The method identifies
significant parameters of the data following different standard feature engineering schemes and subsequently implement the XGboost
regression model. The experimental results show that the proposed framework outperforms the state of the arts in terms of mean
absolute error (MAE) and root mean squared error (RMSE). In future, the aim is to explore the performance of this framework to
estimate the visibility of other crucial areas across globe