An Open-Source Tool for Classification Models in Resource-Constrained Hardware

Applications that need to sense, measure, and gather real-time information\nfrom the environment frequently face three main restrictions: power\nconsumption, cost, and lack of infrastructure. Most of the challenges imposed\nby these limitations can be better addressed by embedding Machine Learning (ML)\nclassifiers in the hardware that senses the environment, creating smart sensors\nable to interpret the low-level data stream. However, for this approach to be\ncost-effective, we need highly efficient classifiers suitable to execute in\nunresourceful hardware, such as low-power microcontrollers. In this paper, we\npresent an open-source tool named EmbML - Embedded Machine Learning that\nimplements a pipeline to develop classifiers for resource-constrained hardware.\nWe describe its implementation details and provide a comprehensive analysis of\nits classifiers considering accuracy, classification time, and memory usage.\nMoreover, we compare the performance of its classifiers with classifiers\nproduced by related tools to demonstrate that our tool provides a diverse set\nof classification algorithms that are both compact and accurate. Finally, we\nvalidate EmbML classifiers in a practical application of a smart sensor and\ntrap for disease vector mosquitoes.\n