An Integrated MLOps Framework for Developing, Tracking, and Deploying Ensemble Learning Models for Lower Limb Motion Classification
Abstract: Wearable inertial sensing has become an effective approach for lower limb motion classification, yet many existing studies primarily emphasize classification accuracy while providing limited support for reproducible model development, experiment tracking, and deployment. This paper presents a reproducible Machine Learning Operations (MLOps) framework for lower limb motion classification using biomechanically meaningful lower-limb joint-angle features obtained from wearable inertial measurement unit (IMU) measurements. The framework processes hip, knee, and ankle joint-angle signals, performs sliding-window feature extraction using statistical, derivative based, and frequency-domain descriptors, and trains multiple supervised machine learning models, including XGBoost, Extra Trees, CatBoost, Ridge Classifier, and Histogram Gradient Boosting. The complete machine learning lifecycle is integrated using Git for version control, MLflow for experiment tracking, Docker for reproducible containerized deployment, and Streamlit for interactive inference and visualization. The framework was evaluated using Xsens Awinda data collected from five subjects performing six lower limb activities, namely Walking, Sitting, Standing, Stairs, SlopeUp, and SlopeDown, with 1,617 feature windows used for the reported test evaluation. Experimental results demonstrate that the Extra Trees classifier achieved the highest performance, with an accuracy of 99.59 percent and a Macro F1-score of 0.9939, outperforming the other evaluated machine learning models. The proposed framework enables reproducible model training, artifact management, deployment, and interactive inference within a unified workflow while improving the interpretability of activity recognition through joint angle-based biomechanical feature representation. The proposed MLOps framework provides a practical and scalable solution for developing reproducible wearable sensing applications and offers a foundation for future real-time, multimodal, and edge deployable lower limb motion analysis systems.
Keywords: MLOps, Machine Learning Pipeline, Lower Limb Motion Classification, Joint Kinematics, IMU, MLflow, Docker, Git, Streamlit, Ensemble Learning, Experiment Tracking, Model Deployment
How to Cite?: R. Anitha, Geetesh N., Vikash Kumar, K. Mohanavelu, "An Integrated MLOps Framework for Developing, Tracking, and Deploying Ensemble Learning Models for Lower Limb Motion Classification", Volume 15 Issue 10, October 2026, International Journal of Science and Research (IJSR), Pages: 579-591, https://www.ijsr.net/getabstract.php?paperid=SR26827162518, DOI: https://dx.doi.org/10.21275/SR26827162518