Bitumen is extracted from oil sands using warm water and additives. The resulting bitumen froth is diluted with naphtha in a froth treatment process. Residual naphtha in the aqueous tailings of the froth treatment unit is recovered in a naphtha recovery unit (NRU). It is imperative to maximize the naphtha recovery process to minimize the plant's environmental and economic impact. It is, in this respect, that NRU vapour–liquid–liquid equilibrium data is of significant value. In this work, a paraffinic‐aromatic synthetic naphtha (PASN) with a true boiling point (TBP) similar to that of froth treatment naphtha is used. Water/PASN mixtures are studied using the Soave‐Redlich‐Kwong equation of state with a Kabadi‐Danner modification. The tangent plane distance (TPD) is evaluated as a possible criterion to calculate the number of phases, with its significant shortcomings being established. As well, experimental data obtained in a CREC‐VL‐Cell is observed to display higher solubilities of PASN in water than the ones obtained by HYSYS‐Aspen Plus V9 simulation. To address these issues, a machine learning (ML)‐based phase classification methodology was considered, predicting the number of phases with a 99% recall. This anticipates that ML will be of significant value for faster convergence of the flash split calculations for the naphtha hydrocarbon‐water systems under consideration.