Estimating Plant Physiological State by Learning Methods Considering Imbalance and Time-series
Koki Sato, Naoki Oishi, Natsuru Futamata, Hiroshi Mineno · 2023
This study considered machine learning methods for accurate estimation and prediction of imbalanced and time-series data. We used machine learning to estimate plant physiological states from environmental data with imbalance and time-series. We proposed improvement the issues of CREAMER and a learning method considering time-series. CREAMER, a resampling method that resolves imbalance, had issues in setting hyperparameters, so we selected from multiple candidates. In order to consider time-series, resampling was applied on the dimension-compressed data in 1D-CAE. By restoring this data to its original dimensions, resampling without loss of time-series was possible. As a validation of the CREAMER hyperparameters, the proposed parameters were validated by comparing with the comparison parameters. As a verification of the effectiveness of a learning method considering time-series, we confirmed that it cannot be effective by comparing it with a method not considering time-series.