The gamma-ray logging method is one of the mandatory well logging methods for geophysical exploration of wells. However, during the conduct of such a study, the sensor, for one reason or another, may stop recording observations in the well. If a small number of values are missing, you can restore these values using standard methods to fill in gaps like in time series. If data miss a large number of values, observations usually are made again, which leads to additional financial costs. This work proposes an alternative solution, in the form of filling missed observations in data with the help of machine learning methods. The main idea of this method is to construct a simple two-layer neural network that is trained on data from the well, and then synthesise the missing values based on the trained neural network. This work evaluates the effectiveness of the proposed method, and gives reasons for the appropriateness of using different methods of filling gaps, depending on the number of missed values.