The article discusses the possibilities of predicting the state of the web resources usability.  Next, modeling and predicting system the time series of the site page visitors number is proposed. The usability testing procedure is quite costly from both financial and time points of view. Thus, a system that reduces these costs is useful for modern organizations. Different approaches of forecasting the number of visitors: ARIMA model and Neural Networks are considered. An important time series property for ARIMA model being applicable is the stationarity of the series. It is shown that this model is not suitable enough for the investigated time series, some types of neural networks are also not suitable for various reasons. As a result, NARX networks are selected, which are successfully used for time series forecasting, providing an opportunity to use an exogenous variable. Proposed model using the NARX neural network. The developed application allows evaluating the usability of information resources according to the following parameters: page functionality, accessibility (convenience of location) of the main buttons on the page, colors, type and size of headings and body text. The application helps to make a predict of web resource attendance based on a time series of site visits without usability testing by experts, by comparing simulation results with the results of a full-fledged usability test.