In order to bridge the gap of more than 15m between the drilling bit andhigh-fidelity rock type sensors during the directional drilling, we present anovel approach for identifying rock type at the drilling bit. The approach isbased on application of machine learning techniques for Measurements WhileDrilling (MWD) data. We demonstrate capabilities of the developed approach fordistinguishing between the rock types corresponding to (1) a target oil bearinginterval of a reservoir and (2) a non-productive shale layer and compare it tomore traditional physics-driven approaches. The dataset includes MWD data andlithology mapping along multiple wellbores obtained by processing of LoggingWhile Drilling (LWD) measurements from a massive drilling effort on one of themajor newly developed oilfield in the North of Western Siberia. We comparevarious machine-learning algorithms, examine extra features coming fromphysical modeling of drilling mechanics, and show that the classification errorcan be reduced from 13.5% to 9%.