Implementing Federated Learning based on RainForest Model

Abstract

Federated Learning (FL) is a new concept in machine learning that trains predictive models across multiple nodes and machines holding local training data without sharing that data with other nodes. It’s also known as collaborative learning. In FL, instead of sending the training data, only parameter values are uploaded to the master node or server. On the contrary, Rainforest is a concept for dealing with big data using a decision tree (DT) classifier. DT is one of the popular machine learning algorithms, which is a top-down recursive divide and conquer method. In this paper, we utilized the concept of federated learning by applying the Rainforest algorithm, where datasets are divided into several subsets of data using the clustering technique, from which scalable decision trees are constructed. From each subset of data, we have an AVC (attribute-value, class-label) table, which is sent to the central master node or server to create the full decision tree by using matrix addition. To train the model, we used ten different datasets and evaluated the proposed model. The experimental analysis shows very good accuracy and precision in overall performance.