Performance Analysis of Scientific and Technical Solutions Evaluation Based on Machine Learning

The paper addresses specific results of the study dealing with data analytic system development for scientific and technical solutions and technologies efficiency assessment and provides the comparison of the application results for different algorithms, including Random Forest, Extreme Gradient Boosting (XGBoost) as well as hybrid neural networks. In order to illustrate practical results of the proposed system, the analysis of the effectiveness and feasibility of scientific and technical solutions and technologies was carried out for the Power & Energy Sector of Russia, namely for the innovation projects, dealing with energy efficiency and energy saving and presented at business incubator competitions and/or submitted to state scientific foundations for the period from 2010 ti112017. The paper provides the result of the classification accuracy analysis for the algorithms under consideration. The obtained results characterize the gradient boosting over decision trees as fairly reliable with an accuracy of 89,4%. In addition to the generally accepted metrics, precision and recall parameters were considered to improve the quality of the resulting error analysis.