Resumen
The problem of software failures in the operation of complex corporate software systems is economically significant and, unfortunately, inevitable. Therefore, to solve this problem, it is necessary to predict failures in a timely manner, based on information from the event logs (logs) of individual applications.This paper investigates the capabilities of neural networks for predicting the performance of applications running on technological platforms and detecting anomalous situations. Network training is carried out on the basis of information from the log metrics of individual applications of technological platforms.Were considered linear, dense, convolutional, recurrent neural networks, as well as the Holt-Winters and XGBoost models. The quality metric MAE (mean absolute error) is used to compare the results. As a result of comparisons, the best result was shown by recurrent neural networks (AR LSTM).