Objective. The goal of this study was to develop control strategies to produce alternating, weight-bearing stepping in a cat model of hemisection spinal cord injury (SCI) using intraspinal microstimulation (ISMS). Approach. Six cats were anesthetized and the functional consequences of a hemisection SCI were simulated by manually moving one hind-limb through the gait cycle over a moving treadmill belt. ISMS activated the muscles in the other leg by stimulating motor networks in the lumbosacral enlargement using low levels of current (<110 µA). The control strategy used signals from ground reaction forces and angular velocity from the manually-moved limb to anticipate states of the gait cycle, and controlled ISMS to move the other hind-limb into the opposite state. Adaptive control strategies were developed to ensure weight-bearing at different stepping speeds. The step period was predicted using generalizations obtained through four supervised machine learning algorithms and used to adapt the control strategy for faster steps. Main results. At a single speed, 100% of the steps had sufficient weight-bearing; at faster speeds without adaptation, 97.6% of steps were weight-bearing (significantly less than that for single speed; p = 0.002). By adapting the control strategy for faster steps using the predicted step period, weight-bearing was achieved in more than 99% of the steps in three of four methods (significantly more than without adaptation p < 0.04). Overall, a multivariate model tree increased the number of weight-bearing steps, restored step symmetry, and maintained alternation at faster stepping speeds. Significance. Through the adaptive control strategies guided by supervised machine learning, we were able to restore weight-bearing and maintain alternation and step symmetry at varying stepping speeds.