In a mobile sensing device, a method of detecting blood alcohol content includes receiving time-series gait data from at least one sensor of the mobile sensing device as a user walks and detecting a set of attributes associated with the time-series gait data, each attribute of the set of attributes related to the user's gait. The method includes comparing the set of attributes with a machine learning classification model learned from a training data set of attributes to determine at least one of a blood alcohol content range of the user and an impairment level of the user and outputting a notification associated with the at least one of the blood alcohol content range of the user and the impairment level of the user.