Neural Network Representation Learning for Food Product Clustering Analysis

This project:

• Built Variational Autoencoder (VAE) architecture using Keras framework in Python, made use of its Encoder layer to generate new feature/latent representation of food nutrition dataset from United States Department of Agriculture (USDA).
• Compared with raw representation and PCA representation as input of three clustering algorithms: K-means (centroid-based), Agglomerative Hierarchical Clustering (hierarchical-based), and Gaussian Mixture Model (distribution-based).
• The clustering models can be used for Nutritionist/Dietitian/etc., food producer, or even end-consumer for further analysis in terms of food nutrition.