TinyDecisionTreeClassifier: Embedded C++ library for training and applying decision trees on the edge

Abstract

Machine learning and AI remain hot topics in research. However, most machine-learning models are trained and applied to big data. Major electronics parts manufacturers are actively working on simplifying the deployment of machine learning models to their chips. Software companies provide microcontroller code generation after the training data are uploaded to their platform. Unfortunately, most of the available open-source solutions do not support training models on the edge. The proposed TinyDecisionTreeClassifier is a standalone open-source C++ library that allows both training and deployment on the edge. This paper also covers the deployment and benchmark of the software performance and power consumption on some of the commonly used microcontrollers.

Keywords

• TinyML
• Decision tree
• Arduino
• C++
• Machine learning