This study authenticated fish feed sources and determined lard adulteration using dataset pre-processing, principal component analysis (PCA), discriminant analysis (DA) and partial least square regression (PLSR) on 19 triacylglycerols (TAGs) and 16 thermal properties (TPs). At cumulative variability (90.625%) and Keiser-Meyer Olkin (KMO) value (0.811), the PCA identified strong factor loading variables, i.e., OLL, PLL, OOL, POL, PPL, POO, PPO, PSO, ICT and FHT in PC1 and LLLn, OOO and CT2 in PC2. These variables were significantly (p < 0.05) contributing to lard-palm-oil (L-PO) clusters: (1) POO, PPO and PPL (high loading) and OLL, PLL, OOL, ICT, POL, PSO and FHT (low loading) in 0:100 and 25:75 L-PO clusters; (2) CT2, OOO and LLLn (high loading) in 50:50 L-PO cluster; and (3) OLL, PLL, OOL, ICT, POL, PSO and FHT (high loading) and POO, PPO and PPL (low loading) in 72:25 and 100:0 L-PO clusters. Training, validation and testing datasets had 100%, 84.44% and 100% correct-classification, respectively at p < 0.0001 of Wilks' lambda and p < 0.0001 Fisher distance. The DA selected PLL, OOL, POL, PPL, PSO, ICT and FHT as the significantly authenticating biomarkers (p < 0.05). With determination coefficient (R²) (0.9693), mean square error (MSE) (38.382) and root mean square error (RMSE) (6.195), the PLSR's variable importance in the projection (VIP) identified the most influential biomarkers, i.e., PPL, POL, PPO, OOL, ICT, PLL, FHT, POO and OLL. The Z-test result (p > 0.05) indicated that the PLSR could determine the lard adulteration percentage in fish feed.