Human Activity Recognition Using Binary Sensors, BLE Beacons, an Intelligent Floor and Acceleration Data: A Machine Learning Approach

Abstract

Although there have been many studies aimed at the field of Human Activity Recognition, the relationship between what we do and where we do it has been little explored in this field. The objective of this paper is to propose an approach based on machine learning to address the challenge of the 1st UCAmI cup, which is the recognition of 24 activities of daily living using a dataset that allows to explore the aforementioned relationship, since it contains data collected from four data sources: binary sensors, an intelligent floor, proximity and acceleration sensors. The methodology for data mining projects CRISP-DM was followed in this work. To perform synchronization and classification tasks a java desktop application was developed. As a result, the accuracy achieved in the classification of the 24 activities using 10-fold-cross-validation on the training dataset was 92.1%, but an accuracy of 60.1% was obtained on the test dataset. The low accuracy of the classification might be caused by the class imbalance of the training dataset; therefore, more labeled data are necessary for training the algorithm. Although we could not obtain an optimal result, it is possible to iterate in the methodology to look for a way to improve the obtained results.

UJA Human Activity Recognition multi-occupancy dataset

“…In the work [13] This dataset integrated multiple activities of a single inhabitant in a smart lab at the University of Jaén (Spain) with a heterogeneous set of devices with sensors. The most relevant proposals were based on the techniques: bagging classifier [14], finite state machine [15], filtered classifier [16], finite automata and regular expressions [17], naive Bayes classifier [18] and hidden Markov model + definition-based model [19].…”

“…In [14], the dataset was processed using six classification methods (Decision Tree (C4.5), 1 Nearest-Neighbor (1-NN), Support Vector Machine SVM, random forest, AdaBoostM1 and bagging), developing the Cross Industry Standard Process for Data Mining methodology. In this experimentation, the accuracy in the recognition of the 24 activities was 92.10% with an evaluation model based on 10-fold cross validation.…”

Neural Network Ensembles for Sensor-Based Human Activity Recognition Within Smart Environments

“…In [72], the challenge organizers describe the UCAmI Cup dataset comprehensively. Knowledge-driven rule-based approaches outperformed the data-driven approaches to the activity recognition problem, with many of the participants reporting issues and limitations found within the data [73][74][75][76]. The approach implemented by [73] involved a domain knowledge-based solution inspired by a Finite State Machine, achieving 81.3% accuracy.…”

“…Reported results show mean accuracies of around 68%, with the researchers stating that given the high number of activity classes, the outcome achieved was reasonable. Another approach implemented in [76] used various common machine learning algorithms, including a Decision Tree, Nearest Neighbour, Support Vector Machine, and three ensemble approaches including Random Forest, Boosting, and Bagging. The researchers reported a training set accuracy of 92.1%, however, their approach achieved 60.1% on the provided test data which demonstrated poor generalization.…”

“…In [76], UCAmI Cup participants suggested the cause for their low HAR performance was the high imbalance of classes in the training set and stated that the training algorithm required more labelled training data to perform better.…”