eSiebold95/Walking_Classification: Lab-to-Field Transfer of Gait Recognition Models (v1.0.0)

This release contains the code used in the ECSS 2026 abstract "Lab to Field Transfer of Gait Recognition Models", examining the transferability of gait recognition models from laboratory to military field conditions using upper-arm IMU data. Two approaches are implemented: a convolutional neural network trained on z-normalized raw acceleration magnitude, and a frequency-domain model using FFT features with XGBoost. Models are trained on laboratory data and evaluated on laboratory and field data, as well as on combined datasets, using subject-wise cross-validation and randomized hyperparameter search. Laboratory data include treadmill and non-locomotion activities, while field data were collected during a 100 km military march. Raw and processed data are not included; local paths are externalized and excluded from version control.