Hibrid approach to processing incomplete stream data in distributed real-time systems

Abstract

The article considers the problem of processing incomplete streaming data in distributed real-time systems, in particular in the context of data mining. It is noted that traditional methods of imputation are ineffective in conditions of limited resources, high requirements for processing speed and dynamic nature of streams. A hybrid approach combining federated learning, contextual imputation and adaptation to conceptual drift is proposed. The method allows local distributed computing nodes to train lightweight imputation models on their own data, followed by centralised aggregation, backpropagation of the global model and its dynamic updating. Experimental verification on a real dataset has shown the advantages of the approach in terms of accuracy (RMSE, MAE) and network load compared to the baseline methods. The obtained results prove the effectiveness of the proposed method in distributed environments with limited computing resources.

Prombles in programming 2025; 2: 112-121