Ensemble and Deep Learning for Real-time Sensors Evaluation of algorithms for real-time sensors with application for detecting brushing location

Real-time sensors on handheld devices place unique requirements on machine learning methods. In this study, ensemble and deep learning methods usage for real-time sensors are explored. The focus is on learning methods applied in smart toothbrush devices with the potential to improve overall health as dental caries is the most frequent chronical disease for children. For preventing cavities, smart toothbrushes are a valuable tool in guiding the user to which teeth and surface to brush. The core of a smart toothbrush is the algorithm to detect the location brushed in real-time. To improve the knowledge about promising methods for this task, various machine learning algorithms are examined, utilizing the data generated by real-time sensors on the toothbrush for the localization of the same in the mouth of the user. A final dataset of more than one million rows (1.44M), seven features and seventy-two classes is tested with KNN, Naïve Bayes, Support Vector Machines, Extra Trees Classifier and various voting classifiers. The results, compared across various dimensions such as prediction performance, model training, prediction overhead and model size, show that the ExtraTree classifier is the most promising method of the evaluated algorithms with an F1 score of 58.77%.