Identifying effective peptide inhibitors of DPP-IV is vital for Type 2 Diabetes treatment, as they preserve incretin hormones (GLP-1/GIP) and support glycemic control. However, traditional wet-lab discovery methods are often expensive and time-intensive. To overcome this limitation, we developed a computationally efficient machine learning approach based on interpretable amino acid features, including composition, transition, and distribution. Our optimized model achieved robust performance (AUC-ROC: 0.95 \({\pm \,\,\, 0.01}\) ; F1-score: 0.87 \({\pm \,\,\, 0.03}\) ; accuracy: 0.86 \(_{\pm 0.03}\) ) using only 14 key descriptors, matching state-of-the-art methods while offering greater simplicity. Feature analysis revealed that hydrophobicity, neutral charge, N-terminal physicochemical properties, and sequence transition patterns play critical roles in DPP-IV inhibition. These findings provide clear guidelines for the rational design of peptide inhibitors, potentially accelerating the development of novel diabetes therapeutics.