•Total of 198 volatile compounds were identified in five commercial grilled lamb shashliks.•Characterizing commercial shashliks flavor by intelligent sensory technology.•Predicting the concentrations of eight classes of VOC and commercial shashliks brands.•The MDBN models achieved a better predictive performance than other models. HS-SPME-GC-MS, SPME-Arrow-GC × GC-TOF-MS, HS-GC-IMS, Electronic-nose, and Electronic-tongue systems were applied in a feasibility study of the flavor characterization of five commercially available Chinese grilled lamb shashliks. A total of 198 volatile organic compounds (VOCs) were identified (∼71% by GC × GC-TOF-MS). Using data fusion strategies, five predictive models were applied to the composition of VOCs and brand identification of the lamb shashliks. Compared with partial least squares regression, support vector machine, deep neural network, and RegBoost modeling, a momentum deep belief network model performed best in predicting VOCs content and identifying shashlik brands (R2 above 0.96, and RMSE below 0.1). Intelligent sensory technology combined with chemometrics is a promising approach to the flavor characterization of shashliks and other food matrices.