Decision tree learning with random forest models using agricultural and ecological field data incorporating multi-factor studies and covariate structure

Abstract In this study, the decision learning methods of regression tree and random forest analysis are investigated as complements to standard statistical methods such as analysis of variance and grouped regression. For this purpose, three diverse data sets were used. The first set is large and multidimensional and describes nitrous oxide emissions from sites across different geo-positions in the UK receiving various fertilisation treatments. The second set is based on Gliricidia tree provenances and has a small number of samples and an imbalanced distribution of factor classes. Random forest modelling was found to be a very viable option in the case of the first data set but failed in the case of second. The third data set, based on count observations recording osprey egg incubation times, lends itself to tree and forest modelling. These decision learning methods therefore appear well suited to handling the diverse, multi-dimensional and complex data sets that often arise in carrying out agricultural and ecological field experiments.