Abstarct: Accurately predicting enzyme inhibition effect is essential for guiding drug design and discovery and for deepening the understanding of biochemical pathways. In this study, two chemometric modeling approaches RF-ANN (a hybrid method combining Random Forest–baxsed variable selection with an Artificial Neural Network) and FF-ANN (Fire Fly variable selection and modeling using ANN) were developed to predict the inhibitory activity of a series of derivatives targeting α-amino-β-carboxymuconate-ε-semialdehyde decarboxylase (ACMSD), a key enzyme in the kynurenine mextabolic pathway. Relevant molecular descxriptors were selected using stepwise regression, Fire Fly and Random Forest algorithms, and predictive models were subsequently constructed using Artificial Neural Networks. The performance of the RF-ANN and FF-ANN models was evaluated using statistical indicators including R², MSE, and RMSE. The results showed that the RF-ANN model achieved greater predictive accuracy and better generalization capability than the FF-ANN model, effectively capturing nonlinear relationships among the descxriptors. But the number of descxriptors in FF-ANN is lower than RF-ANN and so this model is simpler. The obtained results indicate that the applied modeling strategies exhibit strong predictive power for estimating the experimental inhibitory potency of the studied compounds, yielding R² values of 0.8954 for the RF-ANN model and 0.8231 for the FF-ANN model.