Abnormal Behaviour Detection for Dementia Sufferers via Transfer Learning and Recursive Auto-Encoders

Abstract

Cognitive impairment is one of the crucial problems elderly people face. Tracking their daily life activities and detecting early indicators of cognitive decline would be necessary for further diagnosis. Depending on the decline magnitude, monitoring may need to be done over long periods of time to detect abnormal behaviour. In the absence of training data, it would be helpful to learn the normal behaviour and daily life patterns of a (cognitively) healthy person and use them as a basis for tracking other patients. In this paper, we propose to investigate Recursive Auto-Encoders (RAE)-based transfer learning to cope with the problem of scarcity of data in the context of abnormal behaviour detection. We present a method for generating synthetic data to reflect on some behavior of people with dementia. An RAE model is trained on data of a healthy person in a source household. Then, the resulting RAE is used to detect abnormal behavior in a target house. To evaluate the proposed approach, we compare the results with the-state-ofthe-art supervised methods. The results indicate that transfer learning is promising when there is lack of training data.

Detection of Dementia-Related Abnormal Behaviour Using Recursive Auto-Encoders

“…Automatic assessment of cognitive impairment has been tackled using many machine learning approaches, such as support vector machines (SVMs), naïve Bayes (NB) methods [ 21 ], restricted Boltzmann machines (RBMs) [ 22 ], Markov logic networks [ 9 , 23 ], hidden Markov models (HMMs) [ 19 , 24 ], random forest methods [ 20 ], hidden conditional random fields [ 25 ], recurrent neural networks (RNNs), convolutional neural networks (CNNs) [ 26 , 27 ] and some hierarchical models [ 28 , 29 ].…”

“…In our study, the abnormal behaviour is defined in the context of sequences considering their relationships with before and after activities, similarly to [ 9 , 25 ]. In [ 28 ], recursive auto-encoders (RAE) were used to cope with the scarcity of data. The authors applied transfer learning when there was limited data available.…”

“…Data Simulation Studies: Many studies used data simulations to cope with the scarcity of data [ 17 , 19 , 20 ]. In [ 28 ], transfer learning via recursive auto-encoders (RAE) was used to detect abnormal behaviour of elderly people when there was limited data available. First, normal behaviour was learned in a source household, and then the parameters of a RAE were transferred to another house (target) to detect abnormal behaviour.…”

“…In [ 17 ], the authors modified a real-life dataset of an older adult converting basically the rooms into activities. In [ 28 ], recursive auto-encoders (RAE) via transfer learning were used to cope with the scarcity of data. First “normal” behaviour in a source household was learnt, and then the parameters of a RAE were transferred to another house (source) to detect abnormal behaviour of dementia sufferers.…”

Detecting indicators of cognitive impairment via Graph Convolutional Networks

“…Recent studies show that deviations in activity patterns can be indicators of cognitive decline [2,3,4,5]. Although people don't perform the same activities in the same way every time, these activities still follow a set of patterns in terms of locations, time and frequency in each day.…”