Участник:Aignatov

Материал из MachineLearning.

Версия от 17:52, 30 декабря 2014; Aignatov (Обсуждение | вклад)
(разн.) ← Предыдущая | Текущая версия (разн.) | Следующая → (разн.)
Перейти к: навигация, поиск

Игнатов Андрей Дмитриевич

МФТИ, ФУПМ, 174

andrey.ignatoff@gmail.com

Отчеты о научно-исследовательской работе

Весна 2014, 6-й семестр

Human activity types recognition using quasiperiodic sets of time series

This paper investigates the classification problem of time series, received from the accelerometer of mobile phone. Primarily it solves the problem of time series segmentation, assuming that each segment corresponds to one fundamental period of motion. The obtained segments refer to various types of human physical activity. One must to detect a type for each segment. The principal component analysis is applied to extract the fundamental period and to remove noise from the data. To recognize activities we use k-nearest neighbor method and neural network as alternative. We verify accuracy of obtained algorithms by testing them on the sample of measured dataset and compare proposed methods.

Публикация


Осень 2014, 7-й семестр

Testing ground for classification algorithms

The majority of machine learning problems can be solved using various classification techniques, and the selection of an appropriate one is a quite complicated task. This project is performed in order to create a web-based platform for quick performance evaluation of different classification algorithms that can be applied to a particular problem. The platform requires the problem data to be of the form of the object-feature matrix. Since the obtained classification results must be comparable, we provide the identical testing method for the embedded algorithms that is based on the cross validation technique. The developed client-server web application is available at: http://remote.vdi.mipt.ru:60080.

Технический отчет

Доклад на конференции

  • 57-ая конференция МФТИ, доклад "Распознавание типа физической активности человека по данным, полученным с акселерометра мобильного телефона"
Личные инструменты