Evaluation of fall detection classification approaches

Abstract

As we grow old, our desire for being independence does not decrease while our health needs to be monitored more frequently. Accidents such as falling can be a serious problem for the elderly. An accurate automatic fall detection system can help elderly people be safe in every situation. In this paper a waist worn fall detection system has been proposed. A tri-axial accelerometer (ADXL345) was used to capture the movement signals of human body and detect events such as walking and falling to a reasonable degree of accuracy. A set of laboratory-based falls and activities of daily living (ADL) were performed by healthy volunteers with different physical characteristics. This paper presents the comparison of different machine learning classification algorithms using Waikato Environment for Knowledge Analysis (WEKA) platform for classifying falling patterns from ADL patterns. The aim of this paper is to investigate the performance of different classification algorithms for a set of recorded acceleration data. The algorithms are Multilayer Perceptron, Naive Bayes, Decision tree, Support Vector Machine, ZeroR and OneR. The acceleration data with a total data of 6962 instances and 29 attributes were used to evaluate the performance of the different classification algorithm. Results show that the Multilayer Perceptron algorithm is the best option among other mentioned algorithms, due to its high accuracy in fall detection. 978-1-4577-1967-7/12/$26.00 ©2011 IEEE [ 131 ]

Combining novelty detectors to improve accelerometer-based fall detection

“…In addition, the proposed procedure is based on a more principled approach by using a clear distinction between training and validation data sets and a single measure of performance, while many previous algorithms based on simple threshold algorithms found them ad hoc in all the data set [6,15]. Other papers considered several features at the same time, but without any previous feature selection processing to reduce their number [1,16]. However, adding features did not help to improve SVM performance.…”

“…In [16], a windowing technique was used to extract features (1-s sliding window with 0.5-s overlapping). Several features were extracted up to a total of 28 (acceleration, velocity, position and different time domain features).…”

Development of Wearable Human Fall Detection System Using Multilayer Perceptron Neural Network

“…They provide a general and practical method for learning real-valued, discrete-valued and vector-valued from linear and non-linear functions. 18 As explained in our previous work 19 , a multilayer perceptron (MLP) neural network as a commonly used ANN structure was selected as the most powerful classification algorithm for precise classification of motions and determination of fall or ADL events. It is a three-layer ANN consisting of input layer, hidden layer and output layer.…”

Investigation of Sensor Placement for Accurate Fall Detection

“…After pre-processing, features from the time or spatial domain are extracted to feed trained classifiers such as artificial neural (ANN) or Bayesian networks (BN), support vector machines (SVMs), decision trees, k-nearest neighbors (k-NN), etc. Kerdegari et al [12] used statistical features such as maximum, minimum, mean, range, variance and standard deviation extracted from a waist-worn tri-axial accelerometer to investigate the performance of various classifiers on fall detection. The multilayer perceptron yielded the best sensitivity (90.15%).…”