[A-CG43] Earth & Environmental Sciences and Artificial Intelligence/Machine Learning

[ACG43-P05] A Preliminary Study on Developing a Smartphone Application for Weather Classification Using AutoML

*Satoshi Ono¹, Kento Wakamatsu¹, Yusaku Hayashi¹, Tomohiko Tomita² (1.Kagoshima Univeristy, 2.Kumamoto University)

Keywords:machine learning, artificial intelligence, AutoML, neural architecture search, weather classification

In recent years, the need for local weather observation has increased for various applications. For instance, predicting local weather phenomena such as guerrilla rains and tornadoes requires detailed ground-based observation networks. In addition, recording weather conditions in an individual's environment can be used to treat physical illnesses caused by climate and weather changes, known as weather-related diseases. Therefore, this study develops a smartphone application that discriminates the weather as a tool to realize a local meteorological observation network or to record meteorological factors to patients. This study utilizes an automatic machine learning design technique (AutoML) that include neural architecture search and hyper-parameter optimization to design deep neural network architecture for weather classification and compresses the designed architecture to implement it on smartphones.