Lower limb disability severely impacts gait, thus requiring clinical interventions. Inertial sensor systems offer the potential for objective monitoring and assessment of gait in and out of the clinic. However, it is imperative such systems are capable of measuring important gait parameters while being minimally obtrusive (requiring few sensors). This work used convolutional neural networks to estimate a set of six spatiotemporal and kinematic gait parameters based on raw inertial sensor data. This differs from previous work which either was limited to spatiotemporal parameters or required conventional strap-down integration techniques to estimate kinematic parameters. Additionally, we investigated a data segmentation method which does not rely on gait event detection, further supporting its applicability in real-world settings.Preliminary results demonstrate our model achieved high accuracy on a mix of spatiotemporal and kinematic gait parameters, either meeting or exceeding benchmarks based on literature. We achieved 0.04 ± 0.03 mean absolute error for stance-time symmetry ratio and an absolute error of 4.78 ± 4.78, 4.50 ± 4.33, and 6.47 ± 7.37cm for right and left step length and stride length, respectively. Lastly, errors for knee and hip ranges of motion were 2.31 ± 4.20 and 1.73 ± 1.93°, respectively. The results suggest that machine learning can be a useful tool for long-term monitoring of gait using a single inertial sensor to estimate measures of gait quality.