<p>Treasury market dominates the fixed-income segment, especially in emerging economies such as India. The Treasury rates affect the investment decisions, portfolio management, capital structure, as well as corporate debt structure decisions of the firms. Even after this huge impact, the Treasury rate prediction problems are less explored. This paper investigates the efficacy of ensemble-based machine learning and deep learning models in predicting the behavior of Treasury bond yields. The work has employed six models on six Treasury index datasets with different maturity periods. The result suggests that the Recurrent Neural Network and its variants are best-suited models to work on Treasury bonds. The Gated Recurrent Unit gives the best result among all Recurrent Network variants. The deep learning models give higher accuracy for the Treasury yield with a greater maturity period. In contrast, the ensemble models perform better on the Treasury yield with a smaller maturity period. The findings call for incorporating machine learning-based prediction in the pricing and valuation of securities, investment plans, and debt structure decisions.</p>