E-nose device, data from GC-MS (measured data), and statistical and mathematical analytic techniques like PCA, PLSR, LDA, and ANN was used in this study and then a GEP programing model developed to estimate caffeine content of samples. Various samples of coffee beans were tested, when caffeine was used as the reference data, R2 for the PLSR and ANN models were 0.9577 and 0.9634, respectively. R2 for the LDA model were identical to 0.9714. Additionally, R2 of the PLSR and ANN models for palmitic acid respectively, was reported 0.893 and 0.9388. Caffeine calibration data produced the greatest results for identifying, according to the information gathered, also GEP model R2 was reported 0.9581. •Non-destructive method for classification of coffee bean varieties and quality.•E-nose machine using models and modeling methods.•PLSR, ANN and LDA as a tool for classification of coffee’s species and cultivars.