This repository contains two Jupyter Notebooks focused on machine learning techniques and their application to toy examples and emotion recognition based on facial landmarks.
TP_IntroSupervised_MachineLearning_0part_toy_classification.ipynb demonstrates how to classify data using different machine learning techniques. The data are 2D points sampled from a multivariate Gaussian distribution, and we aim to classify them into 2, 3 or 4 classes.
- Data Generation & Plotting
- Linear Regression with Integer Transformation and OneHotEncoding
- Logistic Regression, LDA, QDA, GNB, and KNN methods
- Decision Boundaries Analysis
- Comparisons in Computational Time and Test Accuracy
The main takeaway from this notebook is the varying performances of different classification methods depending on the structure and complexity of the data.
TP_IntroSupervised_MachineLearning_1part_FEI.ipynb delves into the concept of emotion recognition based on facial landmarks. This notebook uses the FEI dataset to distinguish between neutral and happy emotions.
- Facial Landmarks Extraction
- Generalized Procrustes Analysis
- Feature Extraction & Scaling
- LDA and Cross-Validation Performance
- Hyperparameter Tuning using Cross-Validation
- Collinearity and PCA
- Landmark Selection
From this notebook, we learn about the complexities involved in facial emotion recognition and the benefits of thoughtful feature selection and scaling.