EXPLORING THE APPLICATION OF VERTICAL FEDERATED LEARNING FOR CLASSIFICATION PROBLEMS

Abstract

The primary goal of occupancy detection is to determine whether the room or any specific place is currently in use by some individual or specific item. This capability holds immense potential for managing efficient electricity, heat, and ventilation in large buildings like hospitals, hotels, or industries. The main objective of this research work is to employ a machine learning technique called vertical federated learning for occupancy detection. In this research occupancy detection dataset from the UCI Machine Learning Repository is used. For occupancy detection, six classifiers are used for the prediction of the highest accuracy and deep learning model. The algorithm of vertical federated learning with categorical cross-entropy loss (vFedCCE) is used to deploy the gradient-based optimizer on the clients, instead of the centralized server using the occupancy detection dataset UCI. The proposed model consists of two clients: client1 and client2, both having the same samples with distinct sets of features. In the initial stage, client1 will develop its model by considering its distinctive features and sharing this model with client2. Client2 will utilize both its model and the model received from client1 to make predictions. After independently calculating its gradient values, client2 will then update its model weights accordingly. This collaborative effort aims to improve the overall model performance, as the client2 will incorporate the gradients obtained from the client1 into its model weights and subsequently return these updated weights to client1. After this client1 updates its model weights based on the received gradients from client2. This process continues until the desired results are achieved. It has been observed that the vFedCCE model exhibits low accuracy as compared to the centralized model while preserving computational cost, privacy, and data bandwidth