Human knee kinematics, especially during gait, are an important analysis tool. The current "gold standard" for kinematics measurement is a multi-camera, marker-based motion capture system with 3D position tracking. These systems are accurate but expensive and their use is limited to a confined laboratory environment. High deflection strain gauges (HDSG) are a novel class of sensors that have the potential to measure kinematics and can be inexpensive, low profile, and are not limited to measurements within a calibrated volume. However, many HDSG sensors can have a non-linear and non-monotonic response. This thesis explores using a nanocomposite HDSG sensor system for measuring knee kinematics in walking gait and overcoming the non-monotonic sensor response found in HDSGs through advanced modeling techniques. Nanocomposite HDSG sensors were placed across the knee joint in nine subjects during walking gait at three speeds and three inclines. The piezoresistive response of the sensors was obtained by including the sensors in a simple electrical circuit and recorded using a low-cost microcontroller. The voltage response from the system was used in four models. The first two models included a physics-based log-normal model and statistical functional data analysis model that estimated continuous knee angles. The third model was a discrete linear regression model that estimated the inflection points on the knee flexion/extension cycle. Finally, a machine learning approach helped to predict subject speed and incline of the walking surface. The models showed the sensor has the capability to provide knee kinematic data to a degree of accuracy comparable to similar kinematic sensors. The log-normal model had a 0.45 r-squared and was unsuitable as a stand-alone continuous angle predictor. After running a 10-fold cross validation the functional data analysis (FDA) model had an overall RMSE of 3.4° and could be used to predict the entire knee flexion/extension angle cycle. The discrete linear regression model predicted the inflection points on the knee kinematics graph during each gait cycle with an average RMSE of 1.92° for angle measures and 0.0332 seconds for time measures. In every estimate, the discrete linear regression model performed better than the FDA model at those points. The 10-fold cross validation of the machine learning approach using the discrete voltages could predict the categorical incline 90% of the time and the RMSE for the speed model was 0.23 MPH. The use of a HDSG as a knee kinematics sensor was shown as a viable alternative to existing motion capture technology. In future work, it is recommended that a calibration method be developed that would allow this sensor to be used independent of a motion capture system. With these advancements, this inexpensive and low profile HDSG will advance understanding of human gait and kinematics in a more affordable and scope enhancing way.