This paper describes the process of developing machine learning models for predicting problem states. The formation of decision support systems in problem situations is based on the using of ensemble methods of machine learning: bagging, boosting and stacking. The algorithms of undersampling and oversampling is applied for improving the quality of the models. Using of complex machine learning models reduces the ability to explain the result obtained, therefore various ways of interpreting the constructed models are given. Based on the results of the study, a method for predicting problem states was formed. This approach contributes to the gradual solution of the identified problem situation and the consistent achievement of the goal.

Keywords: machine learning, bagging, boosting, stacking, problem states, data balancing, shap-values

With the development of wearable technology, unique opportunities have emerged for providing user interaction and highly accurate personalized recognition of his work activity. The purpose of this study is to propose methods for taking into account the evaluation of the economic efficiency of human resources using big data and machine learning methods, which will allow making more informed decisions in the process of human capital management. The article proposes an approach using a hybrid neural network CNN-LSTM, aimed at determining the specific type of work performed by specialists, providing the ability to control the execution of these actions based on data from wearable devices (smart watches, smart bracelets). The accuracy of the developed algorithm in recognizing 18 different types of actions on the test sample was more than 90% according to the Accuracy metric (the proportion of correct answers).

Keywords: human capital, labor productivity, hybrid neural network, convolutional neural network, recurrent neural network