Belkacem Chikhaoui

Clustering home activity distributions for automatic detection of mild cognitive impairment in older adults 1

The public health implications of growing numbers of older adults at risk for dementia places pressure on identifying dementia at its earliest stages so as to develop proactive management plans. The prodromal dementia phase commonly identified as mild cognitive impairment is an important target for this early detection of impending dementia amenable to treatment. In this paper, we propose a method for home-based automatic detection of mild cognitive impairment in older adults through continuous monitoring via unobtrusive sensing technologies. Our method is composed of two main stages: a training stage and a test stage. For training, room activity distributions are estimated for each subject using a time frame of ω weeks, and then affinity propagation is employed to cluster the activity distributions and to extract exemplars to represent the different emerging clusters. For testing, room activity distributions belonging to a test subject with unknown cognitive status are compared to the extracted exemplars and get assigned the labels of the exemplars that result in the smallest normalized Kullbak-Leibler divergence. The labels of the activity distributions are then used to determine the cognitive status of the test subject. Using the sensor and clinical data pertaining to 85 homes with single occupants, we were able to automatically detect mild cognitive impairment in older adults with an F0.5 score of 0.856. Also, we were able to detect the non-amnestic sub-type of mild cognitive impairment in older adults with an F0.5 score of 0.958.

Aggressive and agitated behavior recognition from accelerometer data using non-negative matrix factorization

This paper presents a novel approach for aggressive and agitated behavior recognition using accelerometer data. Our approach applies first a noise reduction technique using the moving average filter method. Then, multiple features such as mean, variance, entropy, correlation and covariance are extracted from the filtered acceleration data using a sliding window. Non-negative matrix factorization is then used in order to project the data into a new reduced space that captures the significant structure of the data. The recognition is performed using the rotation forest ensemble method. The proposed approach is validated using extensive experiments on a real dataset collected at Toronto Rehabilitation Institute. We empirically demonstrate that our proposed approach accurately discriminates between behaviors and performs better than several state-of-the-art approaches.

Ensemble Learning-Based Algorithms for Aggressive and Agitated Behavior Recognition

This paper addresses a practical and challenging problem concerning the recognition of behavioral symptoms dementia (BSD) such as aggressive and agitated behaviors. We propose two new algorithms for the recognition of these behaviors using two different sensors such as a Microsoft Kinect and an Accelerometer sensor. The first algorithm extracts skeleton based features from 3D joint positions data collected by a Kinect sensor, while the second algorithm extracts features from acceleration data collected by a Shimmer accelerometer sensor. Classification is then performed in both algorithms using ensemble learning classifier. We compared the performance of both algorithms in terms of recognition accuracy and processing time. The results obtained, through extensive experiments on a real dataset, showed better performance of the Accelerometer-based algorithm over the Kinect-based algorithm in terms of processing time, and less performance in terms of recognition accuracy. The results also showed how our algorithms outperformed several state of the art methods.

Feature-level combination of skeleton joints and body parts for accurate aggressive and agitated behavior recognition

This paper presents a novel and practical approach for aggressive and agitated behavior recognition using skeleton data. Our approach is based on feature-level combination of joint-based features and body part-based features. To characterize spatiotemporal information, our approach extracts first meaningful joint-based features by computing pairwise distances of skeleton 3D joint positions at each time frame. Then, distances between body parts as well as joint angles are computed to incorporate body part features. These features are then effectively combined using an ensemble learning method based on rotation forests. A singular value decomposition method is used for feature selection and dimensionality reduction. The proposed approach is validated using extensive experiments on variety of challenging 3D action datasets for human behavior recognition. We empirically demonstrate that our proposed approach accurately discriminates between behaviors and performs better than several state of the art algorithms.

A CNN Based Transfer Learning Model for Automatic Activity Recognition from Accelerometer Sensors

Accelerometers are become ubiquitous and available in several devices such as smartphones, smartwaches, fitness trackers, and wearable devices. Accelerometers are increasingly used to monitor human activities of daily living in different contexts such as monitoring activities of persons with cognitive deficits in smart homes, and monitoring physical and fitness activities. Activity recognition is the most important core component in monitoring applications. Activity recognition algorithms require substantial amount of labeled data to produce satisfactory results under diverse circumstances. Several methods have been proposed for activity recognition from accelerometer data. However, very little work has been done on identifying connections and relationships between existing labeled datasets to perform transfer learning for new datasets. In this paper, we investigate deep learning based transfer learning algorithm based on convolutional neural networks (CNNs) that takes advantage of learned representations of activities of daily living from one dataset to recognize these activities in different other datasets characterized by different features including sensor modality, sampling rate, activity duration and environment. We experimentally validated our proposed algorithm on several existing datasets and demonstrated its performance and suitability for activity recognition.

Automatic Identification of Behavior Patterns in Mild Cognitive Impairments and Alzheimer's Disease Based on Activities of Daily Living

The growing number of older adults worldwide places high pressure on identifying dementia at its earliest stages so that early management and intervention strategies could be planned. In this study, we proposed a machine learning based method for automatic identification of behavioral patterns of people with mild cognitive impairment (MCI) and Alzheimer’s disease (AD) through the analysis of data related to their activities of daily living (ADL) collected in two smart home environments. Our method employs first a feature selection technique to extract relevant features for classification and reduce the dimensionality of the data. Then, the output of the feature selection is fed into a random forest classifier for classification. We recruited three groups of participants in our study: healthy older adults, older adults with mild cognitive impairment and older adults with Alzheimer’s disease. We conducted extensive experiments to validate our proposed method. We experimentally showed that our method outperforms state-of-the-art machine learning algorithms.

Challenges in Collecting Big Data in A Clinical Environment with Vulnerable Population: Lessons Learned from A Study Using A Multi-modal Sensors Platform

Agitation is one of the most common behavioural and psychological symptoms in people living with dementia (PLwD). This behaviour can cause tremendous stress and anxiety on family caregivers and healthcare providers. Direct observation of PLwD is the traditional way to measure episodes of agitation. However, this method is subjective, bias-prone and timeconsuming. Importantly, it does not predict the onset of the agitation. Therefore, there is a need to develop a continuous monitoring system that can detect and/or predict the onset of agitation. In this study, a multi-modal sensor platform with video cameras, motion and door sensors, wristbands and pressure mats were set up in a hospital-based dementia behavioural care unit to develop a predictive system to identify the onset of agitation. The research team faced several barriers in the development and initiation of the study, namely addressing concerns about the study ethics, logistics and costs of study activities, device design for PLwD and limitations of its use in the hospital. In this paper, the strategies and methodologies that were implemented to address these challenges are discussed for consideration by future researchers who will conduct similar studies in a hospital setting.

Detecting Communities of Authority and Analyzing Their Influence in Dynamic Social Networks

Users in real-world social networks are organized into communities that differ from each other in terms of influence, authority, interest, size, etc. This article addresses the problems of detecting communities of authority and of estimating the influence of such communities in dynamic social networks. These are new issues that have not yet been addressed in the literature, and they are important in applications such as marketing and recommender systems. To facilitate the identification of communities of authority, our approach first detects communities sharing common interests, which we call “meta-communities,” by incorporating topic modeling based on users’ community memberships. Then, communities of authority are extracted with respect to each meta-community, using a new measure based on the betweenness centrality. To assess the influence between communities over time, we propose a new model based on the Granger causality method. Through extensive experiments on a variety of social network datasets, we empirically demonstrate the suitability of our approach for community-of-authority detection and assessment of the influence between communities over time.