Heart Rate Variability LSTM Autoencoder for Autonomic Fatigue Profiling in Endurance Athletes from Cundinamarca, Colombia: Code, Models and Data

This repository contains the supplementary materials for the doctoral thesis "Autonomic profiling of muscular fatigue in endurance athletes from Cundinamarca using Heart Rate Variability and a Deep LSTM Autoencoder" (Universidad Manuela Beltrán, Bogotá, 2026). CONTENTS:- Phase 2 pipeline: synthetic data generation (CTGAN), LSTM Autoencoder training, 5-fold cross-validation, and latent space interpretability analysis. Includes the trained autoencoder (.h5), encoder (.h5), and fitted StandardScaler (.pkl).- Phase 3 pipeline: application of the pre-trained encoder to 50 endurance athletes from Cundinamarca, K-Means clustering, t-SNE/PCA visualization, Mann-Whitney U tests with Bonferroni correction, and bootstrap-based cluster stability validation.- Jupyter notebooks (Fase II.ipynb, Fase III.ipynb), 24 result tables (.xlsx), 17 publication-quality figures (300 DPI), validated synthetic dataset (4,750 records × 18 HRV variables), and 8-dimensional embeddings. INSTRUMENTATION: Heart Rate Variability was measured using a 15-channel ECG (EDAN SE-15). REPRODUCIBILITY: Random seed SEED=42 fixed across NumPy, TensorFlow, scikit-learn, and CTGAN. Library versions pinned (sdv==1.17.0, scikit-learn==1.5.2). Environment: Google Colab, Python 3.10. Eighteen HRV variables were analyzed: HR, RR mean/max/min, Max/Min ratio, SDNN, RMSSD, NN50, pNN50, SDSD, TINN, triangular index, LF, HF, LF_norm, HF_norm, LF/HF ratio, and total power.