Warp tension is crucial in producing high-quality fabrics and maintaining high loom efficiency. This study aims to mathematically analyze the tension on various parts of the loom while using machine learning (ML) algorithms to predict warp tension. Simple linear regression models (y=α+βx) are used to demonstrate the impact of backrest height and dropper height on warp tension, with r2-value of up to 1.0. In contrast, the retreating backrest roller reveals two key phenomena: the constant distance between the backrest to the dropper and the dropper to the heald frame; both scenarios generate quadratic models (y = ax2 + bx + c) with r2 = 0.97. Force resolution, lever laws, and shed geometry were applied to develop some mathematical equations. Machine learning evaluation confirmed that KMeans clustering effectively categorize the data into 3 groups, while Random Forest algorithm proved most robust for both predicting and classifying warp tension dynamics.