Accurate estimation of joint angles and ground reaction forces/moments (GRF/GRM) is essential for gait analysis, rehabilitation, and assistive device control, but often depends on expensive laboratory equipment. We propose BioKFusion-Net, a deep learning framework that simultaneously estimates GRF/GRM and joint angles from wearable inertial measurement unit (IMU) data. The architecture integrates two task-specific branches with a residual fusion module that captures biomechanical correlations between kinematics and kinetics. Experiments on six healthy subjects show high accuracy in joint angle ( \(R^2 = 0.966\) , NRMSE = 0.027) and GRF/GRM prediction ( \(R^2 = 0.834\) , NRMSE = 0.046), with a 3.1% improvement in force estimation due to the fusion module. This approach enables accurate, portable gait assessment without reliance on motion capture or force plates.