Currently, the issue of correct and timely assessment of the technical condition of wells is quite acute due to the development of fields at a late stage, high water cut of the well product, as well as the aging well stock. Traditionally, geophysical surveys are used to determine the technical condition of the production string, which make it possible to determine the presence of a disturbance, as well as its interval. However, due to the heavy workload of field technicians, it is not always possible to send a geophysical batch in a timely manner to check the technical condition of the well string. This fact entails oil shortages, an increase in the percentage of water cut, a negative impact on the environment, an increase in unproductive injection, as well as a decrease in target economic indicators. It is also worth noting the high risk of "idle" travel, when it is not possible to detect the presence of violations in the production casing.As a solution to these problems, the authors proposed a new technique for assessing the technical condition of the column based on a machine learning model. Chemical analysis of the water, the age of the well, the number of repairs carried out on the well, the dynamics of well operation, as well as its structure are the main features for which well leakage is predicted. The obtained signs are processed, loaded into a machine learning model, and then a specialist is given a real-time conclusion about the presence of a well leak. As a result of the experiments, it was determined that the CatBoost model on the sample, taking into account the class balance, showed the best quality according to the F1-measure metric. The authors interpreted the obtained results of model prediction based on the effect of individual features on the probability of leakage of the production casing of the well both in the whole well stock and for each case separately. The following features have the greatest impact on the condition of the string: well age, sulfate content, specific gravity of water, as well as primary salts.This approach can significantly reduce decision-making time, increase the efficiency of detecting leaks in wells, and increase economic targets.The result of this work is a methodology for detecting leakage of the production casing, as well as an indirect method for diagnosing and predicting the degree of fatigue of the production casing and a certain share of the risk of leakage.