Comparative analysis of machine learning techniques implemented in low cost hardware.
Due to the ease of access to high performance hardwares there has been an exponential growth in the use of algorithms and machine learning techniques in recent years. In parallel to this, Internet access has popularized and the proof of this is that we are surrounded by smart devices such as televisions, cell phones, personal assistants and other portable devices, such as watches, for example. In addition, several low-cost industrial equipment with simple hardware for specific applications, such as environmental monitoring devices, remote controls, and cameras, start to be equipped with Internet access, which allows the use of machine learning algorithms, running in an external server or cluster (in the cloud), to analyze the obtained data . With this scenario, it is useful to know what are the restrictions for running machine learning algorithms on low-cost hardware, as well as the respective performances. This is necessary as it is not always possible to process the data collected by these devices in equipments with greater processing power. Hence, in this work an evaluation of some machine learning algorithms implemented on a set of commonly used low-cost hardware is done, trying to determine the needs of these algorithms with respect to memory requirements and making comparative analysis of the use of each algorithm within each hardware.