An example method includes receiving raw data sets containing drilling parameter and operating condition values generated during subterranean drilling operations. The raw data sets may be separated into training data sets based, at least in part, on the types of the subterranean drilling operations. At least one predictive model may be generated based, at least in part, on at least one training data set. The at least one predictive model may determine a rate of penetration (ROP) for a drilling operation of the same type to which the at least one training data set corresponds.