Autoencoder-Enhanced Roommate Recommendation System

Abstract

This proposed system is designed for creating a new way of giving personalized recommendations by focusing on people's behaviors and preferences. The system uses traditional machine learning algorithms integrating it with deep learning techniques to get the most out of data and suggest recommendations which are designed according to each individual's unique preferences. A deep learning autoencoder is used to learn a lower-dimensional representation of the data, with an assurance on feature extraction and reconstruction accuracy. The encoded features are passed to be clustered using KMeans, with the effectiveness of clustering estimated through internal validation metrics such as silhouette score, Calinski-Harabasz index, and Davies-Bouldin index. Also, t-Distributed Stochastic Neighbor Embedding (t-SNE) is utilized for visualizing clustered data in a simple manner. Additionally, a silhouette plot is given to provide a visual representation of the silhouette scores across clusters, highlighting the degree of cohesion within clusters and separation between them. Finally, using cosine similarity that identifies similar users within the same cluster.