Methodology and validation for identifying gait type using machine learning on IMU data.

Area of Science:

• Biomechanics
• Wearable Technology
• Machine Learning

Background:

• Activity trackers are increasingly popular for monitoring physical activity.
• Distinguishing between different gait types (walking, running) is crucial for personalized rehabilitation and athletic training.
• Current technology often lacks the precision to differentiate gait types effectively.

Purpose of the Study:

• To develop and validate a method for accurately classifying stride types (walking, jogging, running) using wearable inertial measurement units (IMUs).
• To evaluate the performance of artificial neural network (ANN) models in gait type classification based on sensor data.
• To determine the optimal sensor data (acceleration, angular velocity, or both) for accurate gait classification.

Main Methods:

• Fourteen healthy adults performed walking, jogging, and running on a treadmill at controlled speeds.
• An IMU, containing accelerometers and gyroscopes, was attached to each participant's ankle.
• Collected sensor data were processed into time-normalized strides and used to train and test various ANN models.

Main Results:

• The ANN models successfully classified stride types with high accuracy.
• Models utilizing primarily acceleration data achieved the highest accuracy, exceeding 94% after cross-validation.
• The classification performance demonstrated the feasibility of using a single wearable IMU for gait type identification.

Conclusions:

• Artificial neural networks, particularly those trained with acceleration data from a single IMU, can accurately classify different human gait types.
• This technology holds significant potential for enhancing activity tracking, rehabilitation programs, and athletic performance monitoring.
• Further improvements in accuracy are anticipated with the expansion of training datasets for the ANN models.