Classification of the aroma quality of pyrazines derivatives using random forest tree technique

Abstract

We present an alternative classification of the odour molecules based on 0D, 1D, 2D and 3D molecular descriptors by using the Random Forest Tree (RFT). 98 molecules of pyrazine derivatives are classified among three classes of aroma notes: Green, Nutty and Bell-Pepper. The classification model uses 180, 40, 45 and 50 trees in the forest respectively for the 0D, 1D, 2D and 3D descriptors. The use of descriptors 0D, 1D, 2D and 3D correctly classify 72.1%, 70.6%, 82.4% and 85.3% of the molecules during the learning phase. For the test phase, the classification rate is 80%, 86.7%, 93.3% and 90%. This show that RFT is able to develop the molecular Structure - Odour.