At the late stage of field development, residual oil reserves undergo a significant change from mobile to sedentary and stationary. These reserves are mainly located in technogenically and production altered, watered layers and areas of deposits. Localization and development of such sources of hydrocarbons is an effective method of increasing the final oil recovery factor in mature fields, due to the presence of a ready-made developed infrastructure for production, transportation and refining, as well as the availability of highly qualified personnel. This article considers an approach that allows, based on neural network algorithms, the estimation the volumes and localization of residual oil reserves in multi-layer deposits in combination with the analysis of geochemical studies of reservoir fluids. The use of machine learning algorithms allows a targeted approach to the development of residual reserves by automated selection of wellwork. This approach significantly reduces the manual labor of specialists for data processing and decision-making time.