I. Introduction

Knowledge of current weather conditions is of importance in numerous modem applications. Autonomous vehicles are one example where it is crucial to correctly identify the weather condition in order to ensure the passengers’ safety. Agriculture using Internet of Things may also use information from the weather conditions to regulate their irrigation system The use of rain gauges and weather radars are current conventional methods for rainfall detection and estimation. Recently the use of data from commercial microwave links as input to weather classifiers has become increasingly popular due to its high performance. Ericsson captures data from a network of microwave links located in Europa, where the data includes the transmitted and received signal power for each microwave link [1]. The attenuation (deviation from the expected received signal power) may be caused by environmental effects such as rain and snow. Combining the attenuation with artificial intelligence techniques therefore enables performing weather classification. However, weather classification problems generally suffer from imbalanced data which has a direct impact on the model’s accuracy. Currently, a one-dimensional convolutional neural network is used at Ericsson [2], which is not designed for imbalanced learning but has shown good performance on time series problems. Adapting the model to imbalanced learning might detect and solve problems in the industry more efficiently.