This project analyzes the UCI Human Activity Recognition (HAR) dataset and compares two approaches for activity classification:
- Support Vector Machines (SVM), including PCA/LDA dimensionality reduction and hyperparameter tuning.
- A fully connected neural network implemented with TensorFlow/Keras.
The workflow includes exploratory data analysis, subject-aware train/validation splitting, 5-fold StratifiedGroupKFold cross-validation, model selection, and final evaluation on the test set.
exploratory_analysis.ipynb– exploratory analysis, correlations, PCA and LDA.svm.ipynb– SVM baselines, GridSearchCV, cross-validation and test evaluation.deep_learning_model.ipynb– neural-network architectures, early stopping, cross-validation and test evaluation.relazione.pdf– project report.
Install the dependencies with:
pip install -r requirements.txtThen open the notebooks with Jupyter and run them in order.