Finding and reducing electricity losses is one of the key activities of network organizations to improve financial results. Forecasting based on a large number of criteria and comparing with actual electricity consumption is the preferred way to detect losses. However, this process requires a high degree of automation. Therefore, to solve this problem, this paper considers the use of three machine learning algorithms, as well as a comparison of their effectiveness. The author formed a training sample from the database of one of the districts of electrical networks, and also conducted experiments on the implementation of the following algorithms on it: k-nearest neighbors, linear regression and random forest. To compare the resulting models, the author used such performance indicators as mean square error (MSE), absolute mean error (MAE) and coefficient of determination (R^2). The results of the experiment showed the greatest efficiency of the random forest method in comparison with other considered algorithms.