Binder jetting is a multi-step additive manufacturing process that forms metal powders into 3D shapes using a binder, followed by sintering. This technique offers superior surface quality and cost advantages compared to powder bed fusion methods. However, achieving the desired final geometry within tolerances is challenging due to anisotropic shrinkage and gravity-induced deflections during sintering, affecting process reliability. Relying on trial-and-error approaches for each design is inefficient, making numerical predictions crucial. In this study, we conducted sintering experiments on binder-jetted SS316L parts using an in-situ camera monitoring system. The sintering behavior was analyzed for different geometries, including cubes, bridges, diamonds, and lattices, revealing that debinding and sintering are geometry-dependent. Each shape responded differently to identical sintering conditions, highlighting the need for accurate predictive models. To address this, we employed continuum mechanics-based finite element analysis, incorporating empirical fitting parameters derived from experimental data. The model was calibrated using in-situ monitoring results to enable accurate sintering shrinkage predictions. Furthermore, the model was tested using two different powders at different particle size distributions. For different initial green density conditions and different sintering behavior, sintered density predictions for both powder types are in good agreement with experimental results. Additionally, we explored machine learning techniques to establish sintering shrinkage behavior based on experimental training data. By integrating in-situ experimental observations with numerical modeling, this research provides a comprehensive framework for optimizing binder jetting sintering processes through accurate part-scale sintering predictions. The authors gratefully acknowledge the Scientific and Technological Research Council of Türkiye (TUBITAK) under Grant Number 221M200 and 2224-A program for their financial support.