ОПТИМІЗОВАНА МОДЕЛЬ ПРОГНОЗУВАННЯ ДОСТУПНОСТІ ОБ’ЄКТІВ НА ОСНОВІ ГЛИБИННОГО НАВЧАННЯ І ГЕОПРОСТОРОВИХ ОЗНАК

Context. Today, predicting the availability of objects in spatially distributed systems remains one of the areas of computer science that constantly attracts the attention of researchers. There are many reasons for this. There is an increase in the amount of spatial information. New types of infrastructure networks are emerging, as well as the need for rapid decision-making in changing conditions. At the same time, traditional analysis methods do not always cope with the tasks of processing multidimensional data. This is especially true when it comes to complex or unstable environments. This opens up opportunities for applying deep learning methods that demonstrate high efficiency where classical approaches fail.Objective. The study aims to optimize the model for predicting the availability of objects in spatially distributed systems by defining an efficient deep learning architecture that uses spatial and other infrastructure features to improve prediction accuracy and generalizability.Method. To achieve this goal, deep learning architectures were used, including feed-forward models (FNN), convolutional neural networks (CNN), and recurrent neural networks (RNN, GRU, LSTM). During the modeling, methods of data normalization, training regularization, and a comprehensive system for evaluating the accuracy of forecasts using the mean square error, mean absolute error, and coefficient of determination were used.Results. An optimized architecture of a recurrent neural network was built for the study, which includes a combination of two recurrent layers, Dropout regularization layers, and a fully connected layer. The analysis has shown that the proposed model provides high accuracy in predicting the availability of objects, demonstrating stability over a wide range of spatial data. Comparison of actual and predicted values confirmed the effectiveness of the proposed solution.Conclusions. The proposed approach to building an optimized deep learning model for predicting the availability of objects provides a high level of generalization and accuracy, which creates prerequisites for its use in systems of intelligent decision support in spatially distributed environments.