A Cost Forecasting Model for Innovation Activities in the Industrial Sector of Ukraine

Анотація

The article considers the problem of increasing the reliability of estimation of the cost of innovation activity in the industrial sector of Ukraine. Using as primary data the information of the previous study, the dependence of the costs of innovation activity on a group of factors is analyzed, as a result of which it was decided to build several multi-factor regression models. In order to build this group of models, the least squares method was used. In the course of checking the obtained multivariate models, it was found that each model has internal parameters in which the value of p-values exceeds the limit value, which means that the obtained values of the internal parameters of the model are not significant. Therefore, all multivariate models constructed using the least squares method were eliminated from further research. At the next stage of the research, one-factor regression models were constructed using the least squares method, where the amount of mastered production of new types of products (technological processes) was used as a factor. After sorting out the non-essential models, the remaining ones were compared with respect to their quality characteristics. However, in all univariate models, it was found that the calculated values of the average approximation error exceeded 10%. Therefore, yet again, all models have been eliminated from further research. Due to the impossibility of obtaining a model using the least squares method in the simulation process, it was decided to use machine learning methods with a teacher. Among the methods of machine learning, it was decided to pay attention to the following methods: k-near neighbors, regression trees, and neural network method. Taking into account that these models are not adversely affected by the multi-collinearity between factors, both the data for the models were used raw data without additional transformations. According to the results of the research, it was found that among the models using machine learning methods, the best models were the binary regression decision tree (with a hyperparameter value of max_depth = 3) and the neural network model. When comparing these models, it was found that the model based on the decision tree has a smaller value of the average approximation error, which means that this model can be recommended for use in making management decisions on forecasting the costs of innovation activity of industrial enterprises in the future.

Класифікація

Ідентифікатори

Схожі роботи