Krukovets D. Clustering of inflation components and forecasting using machine learning methods

The introduction reveals the essence and current state of the scientific problem, justifying the choice and relevance of developing a multi¬step algorithm based on clustering algorithms and neural networks for time series forecasting, particularly those with the property of disaggregation into subcomponents. The goal, object, subject, and research methods are defined, highlighting the scientific novelty of the research, theoretical and practical significance of the results, the author's personal contribution, and information about the implementation and testing of the results. Forecasting processes and time series are essential for creating strategies with medium¬ and long-term planning, taking into account various factors and risks for acting proactively. However, creating high¬quality predictive models is a nontrivial task. The application of neural networks in economic analysis and forecasting is a relevant and promising direction. The dissertation focuses on developing a combined multi-step neural network model, combining it with clustering algorithms for a disaggregated dataset to simultaneously identify linear and seasonal features in the series and their interdependent structure.n the series and their interdependent structure.