Evaluation of Odor Prediction Model Performance and Variable Importance according to Various Missing Imputation Methods

The aim of this study is to ascertain the most suitable model for predicting complex odors using odor substance data that has a small number and large missing data. First, we compared removal imputation methods, method imputing was found be more effective. Then, in order recommend model, created total 126 models (missing imputation: single imputation, multiple imputations, K-nearest neighbor imputation; preprocessing: standardization, principal component analysis, partial least square; predictive method: regression, machine learning, deep learning) them R2 mean absolute error (MAE) values. Finally, investigated variable importance best prediction model. results identified as combination multivariate Bayesian ridge method, standardization preprocessing, an extremely randomized tree method. Among compounds, Methyl mercaptan, acetic acid, dimethyl sulfide were important compounds odors.