Ensemble Learning in Investment Appraisal

The work is a continuation of the previous research of the authors. In this paper, the ensemble learning method was chosen from the machine learning methods, the essence of which is to use several models (“weak learners”) to solve the same problem. The main principle of the ensemble learning model is that combining weak learners forms a strong learner, thereby increasing the model’s accuracy. As a base of examples necessary for constructing a model, we used data on investment activities in various regions of Russia. Previously, the initial data were divided into several classes using the method of hierarchical clustering. To eliminate the significant imbalance in the original sample, various classification algorithms were used. The solution to the problem was obtained using the Matlab 2018b package, in which the Classifier Learner module was selected, which automatically achieves the result by various methods of ensemble learning. Among the five considered models, the RUSBoosted Trees method showed the best accuracy. The possibility of classifying regions by this method by assigning them to a predetermined class is shown, an error matrix and an operational characteristic are found. The proposed method can be applied in other management tasks related to the assessment of innovations and investments.