=Paper=
{{Paper
|id=Vol-2559/paper3
|storemode=property
|title=Deep Learning and Machine Learning Techniques for Change Detection in Behavior Monitoring
|pdfUrl=https://ceur-ws.org/Vol-2559/paper3.pdf
|volume=Vol-2559
|authors=Giovanni Diraco,Alessandro Leone,Andrea Caroppo,Pietro Siciliano
|dblpUrl=https://dblp.org/rec/conf/aiia/DiracoLCS19
}}
==Deep Learning and Machine Learning Techniques for Change Detection in Behavior Monitoring==
https://ceur-ws.org/Vol-2559/paper3.pdf
1
Deep Learning and Machine Learning Techniques for
Change Detection in Behavior Monitoring
Giovanni Diraco, Alessandro Leone, Andrea Caroppo and Pietro Siciliano
CNR—National Research Council of Italy, IMM—Institute for Microelectronics and Microsys-
tems, Lecce 73010, Italy
{giovanni.diraco, alessandro.leone, andrea.caroppo}@cnr.it,
pietro.siciliano@le.imm.cnr.it
Abstract. Nowadays, smart living environments are equipped with various
kinds of sensors which enable enhanced assisted living services. The availabil-
ity of huge data volumes coming from heterogeneous sources, together with
emerging of novel artificial intelligence methods for data processing and analy-
sis, yields a wide range of actionable insights with the aim to help older adults
to live independently with minimal supervision and/or support from others. In
this scenario, there is a growing demand for technological solutions to monitor
human activities and physiological parameters in order to early detect abnormal
conditions and unusual behaviors. The aim of this study is to compare state-of-
the-art machine learning and deep learning approaches suitable for detecting
early changes in human behavior. At this purpose, specific synthetic datasets
are generated, which include activities of daily living, home locations and vital
signs. The achieved results demonstrate the superiority of deep-learning tech-
niques over traditional supervised/semi-supervised ones in terms of detection
accuracy and lead-time of prediction.
Keywords: Change prediction; machine learning; deep learning; ambient as-
sisted living; human behavior.
1 Introduction
Frail subjects, such as elderly or disabled people, may be at risk when their health
conditions are amenable to change, as it is quite common in case of chronic condi-
tions. That risk can be reduced by early detecting changes in behavioral and/or physi-
cal state, through sensing and assisted living technologies, nowadays available in
smart-living environments. Such technologies, indeed, are able to collect huge
amounts of data by days, months, and even years, providing important information
useful for early detection of changes. Moreover, early change detection makes it pos-
sible to alert formal/informal caregivers and health-care personnel in advance when
significant changes or anomalies are detected, before critical levels are reached and so
preventing chronic diseases. The huge amounts of heterogeneous data collected by
different devices require automated analysis; thus there is a growing interest in auto-
matic systems for detecting abnormal activities and behaviors in the context of smart
Copyright © 2020 for this paper by its authors. Use permitted under Creative Com-
mons License Attribution 4.0 International (CC BY 4.0).
2
living and elderly monitoring [1]. Health monitoring can benefit from knowledge held
in long-term time series of daily activities and behaviors as well as physiological pa-
rameters [2]. A lot of research has been done in the general area of human behavior
understanding, and more specifically in the area of daily activity/behavior recognition
and classification as normal or abnormal [3, 4]. However, very little work is reported
in the literature regarding the evaluation of machine learning (ML) techniques suita-
ble for data analytics in the context of long-term elderly monitoring in smart living
environments. The purpose of this paper is to investigate the most representative clas-
sical machine learning and deep learning (DL) techniques, by comparing them in
detecting/predicting changes in human behavior.
The rest of this paper is organized as follows. The following Section 2 contains relat-
ed works, some background and state-of-the-art in abnormal activity/behavior detec-
tion, with special attention paid to elderly monitoring through heterogeneous data
collected with multi-sensor systems distributed over indoor environments. Section 3
describes materials and methods used in this study, and provides an overview of the
system architecture, the long-term data generation, and the ML/DL techniques. The
findings are presented and discussed in Section 4 and Section 5, respectively. Finally,
Section 6 draws conclusions and some final remarks.
2 Related Work
Today’s available sensing technologies enable long-term continuous monitoring of
activities of daily living (ADLs) and physiological parameters (e.g., heart rate, respi-
ration rate, blood pressure, etc.) in the home environment. Normally, both wearable
and ambient sensing are used, either alone or combined, as multi-sensor systems.
Wearable motion sensors incorporate low-cost accelerometers, gyroscopes and com-
passes, whereas physiological parameter sensors are based on some kind of skin-
contact biosensors (e.g., heart and respiration rates, blood pressure, electrocardiog-
raphy, etc.) [5]. These sensors need to be attached to a wireless wearable node, carried
or worn by the user, needed to process raw data and to transmit/store detected
events/signals. Although wearable devices have the advantage of being usable any-
where and their detection performance is generally good (if the signal-to-noise ratio is
sufficiently high), nevertheless their usage is extremely limited by battery life time
(shortened by the intensive use of wireless communication and on-board processing,
both high energy-demanding tasks) [6], by the need to remember to wear a device,
and by the discomfort of the device itself.
On the other hand, ambient sensing devices are not intrusive in terms of body ob-
struction, since they require the installation of sensors around the home environment,
such as cameras (monocular/stereo, time-of-flight, Lidar, etc.), microphones, sonars,
pyroelectric infrared (PIR) sensors, radar sensors, and pressure/vibration sensors.
Such solutions, blending into the home environment, are generally well-accepted by
end-users [7].
The learning setups for detecting/predicting behavioral changes can be categorized
into three main categories: supervised, semi-supervised, and unsupervised approach-
Copyright © 2020 for this paper by its authors. Use permitted under Creative Com-
mons License Attribution 4.0 International (CC BY 4.0).
3
es. In the supervised case, abnormalities (i.e., changes) are detected via binary classi-
fication in which both normal and abnormal behaviors (i.e., activity sequences) are
labelled and used to learn a model [8, 9, 10]. This model is then applied on real-life
data in order to classify unlabeled behaviors as normal or abnormal. The problem
with this approach is that abnormal behaviors are extremely rare in practice, and thus
the easiest way to collect them is the laboratory simulation or synthetic generation. In
the semi-supervised case, only one kind of labels, i.e., normal behaviors, are used to
train a one-class classifier [11, 12]. Behaviors that do not comply with the learned
model are labeled as outliers during the testing phase. The advantage here is that nor-
mal behaviors, i.e., real-life data, observed during the execution of common ADLs,
are used to train the semi-supervised model (and not simulated or synthetic data as
needed in the supervised case). The last but not least important category includes
unsupervised classifiers, whose training phase does not need any labeling information
(i.e., neither normal nor abnormal behaviors) and any separation into a training and
testing phase [13]. In unsupervised learning, only a small fraction of the observed
behaviors are assumed to be outliers which exhibit a rather different nature than nor-
mal behaviors. In this case, the PRO is that unsupervised-based detection can easily
adapt to various real-life environmental and user’s conditions (where no labeling in-
formation is available); but the disadvantage is that the unsupervised-based detection
requires a quite large amount of initial observations to be fully operational [14, 15].
3 Materials and Methods
For each category of learning setup, i.e., supervised, semi-supervised, and unsu-
pervised, one ML-based and one DL-based technique are evaluated and compared in
terms of detection performance and prediction lead-time at the varying of both normal
behaviors (NB) and abnormal behaviors (AB). All investigated ML and DL tech-
niques are summarized in Table 1. For that purpose, synthetic datasets are generated
by referring to common ADLs and taking into account how older people perform such
activities at their home environment (i.e., instructions and suggestions provided by
geriatricians and leading researches were taken in careful consideration). The synthet-
ic dataset includes six basic ADLs, four home locations in which these activities usu-
ally take place, and five levels of basic vital signs (i.e., heart and respiratory rates)
associated with the execution of each ADL. The six ADLs are activity of eating (AE),
housekeeping (AH), physical exercise (AP), resting (AR), sleeping (AS), toileting
(AT). The considered home locations (LOCs) are bedroom (BR), kitchen (KI), living
room (LR), toilet (TO). Regarding vital signs, for simplicity and without loss of gen-
erality, the only hear rate signal has been considered by dividing it into five levels
(HRLs): very low (VL) < 50 beats/min, low (LO) ∈ [50–80] beats/min, medium (ME)
∈ [80–95] beats/min, high (HI) ∈ [95–110] beats/min, very high (VH) > 110
beats/min.
The objective of this study is to deeply evaluate the learning techniques reported in
Table 1 by considering abnormal datasets, obtained by including the following per-
turbations:
Copyright © 2020 for this paper by its authors. Use permitted under Creative Com-
mons License Attribution 4.0 International (CC BY 4.0).
4
1. Changing of the starting time of one ore more activities (St). This is a change in the
starting time of an activity, e.g., having breakfast at 9 AM instead of 7 AM as usu-
al.
2. Changing of the duration of one or more activities (Du). This change refers to the
duration of an activity, e.g., resting for 3 hours in the afternoon, instead of 1 hour
as usual.
3. Disappearing of one or more activities (Di). In this case, after the change, one ac-
tivity is no more performed by the user, e.g., having physical exercises in the after-
noon.
4. Swapping of two activities (Sw). After the change, two activities are per-formed in
reverse order, e.g., resting and then housekeeping instead of housekeeping and
resting.
5. Changing the location of one or more activities (Lo). One activity usually per-
formed in a home location (e.g., having breakfast in the kitchen), after the change
is performed in a different location (e.g., having breakfast in bed).
6. Changing in heartrate levels when performing one or more activities (Hr). This is a
change in heartrate during an activity, e.g., changing from a low to a high heartrate
during the resting activity in the afternoon.
Although, the sporadic presence of above mentioned changes is not enough to deter-
mine an abnormal condition, nonetheless a sustained change over days or months in
activities, locations, and heartrate levels may be linked to an AB. Hence, the aim of
this study is to evaluate, the ability of ML and DL techniques in predicting such sus-
tained changes, with the objective to notify caregivers/doctors who can use historical
sensor data to make decisions within the application domain of ambient assisted liv-
ing.
In this study, both normal and abnormal long-term (1-year) datasets are realistical-
ly generated by using a probabilistic model based on Hidden Markov Model (HMM)
and Gaussian process (GP). The evaluation metrics adopted in this study are sensitivi-
ty (SEN) and specificity (SPE), defined as follows:
𝑇𝑃 𝑇𝑁
𝑆𝐸𝑁 = , 𝑆𝑃𝐸 = , (1)
𝑇𝑃+𝐹𝑁 𝑇𝑁+𝐹𝑃
where TP is the number of true positives, FP is the number of false positives, TN is
the number of true negatives, and FN is the number of false negatives.
Table 1. Machine learning (ML) and deep learning (DL) techniques compared in this study.
Category Type Technique
Supervised Machine learning Support vector machine (SVM)
Supervised Deep learning Convolutional neural network (CNN)
Semi-supervised Machine learning One-class support vector machine (OCSVM)
Semi-supervised Deep learning Stacked auto-encoders (SAE)
Unsupervised Machine learning K-means clustering (KM)
Unsupervised Deep learning Convolutional auto-encoder (CAE)
Copyright © 2020 for this paper by its authors. Use permitted under Creative Com-
mons License Attribution 4.0 International (CC BY 4.0).
5
The lead-time of prediction (LTP) is defined as follows: maximum number of days,
before the day at which the change becomes stable, within which the future change
can be predicted with the highest performance, i.e., maximizing TP and TN, and min-
imizing FP and FN. Thus, the higher the lead-time (in number of days), the better is
the overall prediction performance.
3.1 Data Generation
In this study, since HMM is used for data generation (in contrast to other studies
where HMM is used for detection purposes [16, 17]), the probabilistic model should
be able to take into account the influence of circadian rhythms on motivated behaviors
(e.g., sleep, hunger, exercise, etc.) [18]. Although suitability of HMM to model ADLs
is encouraged by previous authors’ findings [19], nonetheless further research is
needed to confirm this choice. The user’s physical state bearing diverse ADLs during
the daily circadian cycle is modelled by using three hidden states, i.e., Tired (T),
Hungry (H), and Energized (E), as depicted in Figure 1. Furthermore, each state can
lead to different activities depending on the time of the day (e.g., the state Tired may
lead to Sleeping activity in the night and to Resting activity in the afternoon). Each
arrow of the graph reported in Figure 1 is associated with a probability parameter,
which determines the probability that one state 𝜋𝑖 follows another state 𝜋𝑖 − 1 , i.e., the
transition probability:
𝑎𝑞𝑟 = 𝑃(𝜋𝑖 = 𝑞│𝜋𝑖 − 1 = 𝑟), (2)
where 𝑞, 𝑟 ∈ {𝑇, 𝐻, 𝐸}. The HMM output is a sequence of triples (𝑎, 𝑏, 𝑐) ∈ 𝐴𝐷𝐿 ×
𝐿𝑂𝐶 × 𝐻𝑅𝐿, with ADL={AE,AH,AP,AR,AS,AT}, LOC={BR,KI,LR,TO}, and
HRL={VL,LO,ME,HI,VI} representing, respectively, all possible ADLs, home loca-
tions, and HR levels as previously discussed.
The temporal dependency of activities generated from hidden states is handled by
subdividing a day into four time intervals and modeling the activities in each time
interval with a dedicated HMM sub-model. For each sub-model 𝑀𝑖 , thus, the first
state being activated starts at a time 𝑇𝑖 modeled as a GP, while the other states within
the same sub-model 𝑀𝑖 start in consecutive time slots whose durations are also mod-
eled as GPs.
The ADL, LOC and HRL signals are, normally, sampled at different rates accord-
ing to the specific variability during day time of each signal. For example, the mini-
mum duration of ADLs is of about 10 min, so it is not useful to sample the ADL sig-
nal at 1 min interval. Nonetheless, a unique sampling rate can be adopted for all
measurements. In this study, the sampling rate of 0.2 sample/min (i.e., 5 min interval
between two samples) is selected for all signals. Each dataset is represented in matrix
form with rows and columns equal to the total amount of observed days (i.e., 365
days) and to the total amount of samples per day (i.e., 288 samples), respectively. For
each dataset, the matrix cells can take 120 different values obtained by combining 6
ADLs, 4 locations, and 5 HR levels. For example, the cell value AE-KI-ME indicates
Copyright © 2020 for this paper by its authors. Use permitted under Creative Com-
mons License Attribution 4.0 International (CC BY 4.0).
6
that the subject is eating her meal in the kitchen and her HR level is medium (i.e.,
between 80 and 95 beats/min). Finally, each 1-year dataset is represented by an image
of 365 × 288 pixels with 120 levels of which an example is reported in Figure 2. Ad-
ditionally, for the sake of understanding, each dataset can be represented by splitting
it into three different images, referring to ADLs (6 levels), locations (4 levels), and
HR (5 levels), as shown in Figure 3.
Furthermore, to assess the ability of ML and DL techniques (reported in Table 1)
to detect behavioral abnormalities and changes, the model parameters (i.e., transition
probabilities, emission probabilities, starting times, and durations) were randomly
perturbed in order to generate various kind of abnormal datasets. Without loss of gen-
erality, each abnormal dataset included only one of the abovementioned changes (i.e.,
St, Du, Di, Sw, Lo, Hr) at a time or pairs of them, taken without repetitions (i.e.,
StDu, StDi, StSw, StLo, StHr, etc.).
In order to evaluate the detection performance of ML and DL techniques in Table
1, the HMM parameters (e.g., transition and emission probabilities, starting time and
duration of activities, etc.) are gradually perturbed between the 90th and 180th day,
by randomly interpolating the parameters of the normal and abnormal models. The
resulting perturbed dataset consists of three parts: the first one, ranging from day 1 to
day 90, is referred to normal behavior; the second one, from day 90 to 180, is charac-
terized by gradual changes, becoming progressively more accentuated; the third one,
starting from day 180, is very different from the initial normal period, the change rate
is low or absent, and the subject’s behavior moves into another stability period. An
abnormal dataset, referred to the St change type, is reported in Figure 4.
Fig. 1. State diagram of the Hidden Markov Model (HMM) used to generate long-term activity
data.
Copyright © 2020 for this paper by its authors. Use permitted under Creative Com-
mons License Attribution 4.0 International (CC BY 4.0).
7
Fig. 2. Example of a normal dataset represented as an image of 365 × 288 pixels and 120 levels
(only used levels are reported in the legend).
Fig. 3. The same normal dataset shown in Figure 4 but represented with different images for (a)
activity of daily living (ADL), (b) home locations (LOCs), and (c) heartrate levels (HRLs).
Copyright © 2020 for this paper by its authors. Use permitted under Creative Com-
mons License Attribution 4.0 International (CC BY 4.0).
8
Fig. 4. Example of an abnormal data set, due to a change in the “Starting time of activity” (St).
The change gradually takes place from the 90th day on. (a) ADL, (b) LOCs, and (c) HRLs.
3.2 Abnormal behavior detection
As already anticipated, the learning problem can be addressed by means of differ-
ent learning setups, depending on the label availability. Correspondingly, there are
three main detection approaches, i.e., supervised, semi-supervised and unsupervised,
and which are taken into account in this study as discussed below.
Supervised detection
Supervised detection is based on learning techniques (i.e., classifiers) requiring ful-
ly labelled data for training. This means that both positive samples (i.e., ABs) and
negative samples (i.e., normal behaviors) must be observed and labelled during the
training phase. However, the two label classes are typically strongly unbalanced,
since abnormal events are extremely rare in contrast to normal patterns that instead
are abundant. As a consequence, not all classification techniques are equally effective
for this situation. In practice, some algorithms are not able to deal with unbalanced
data, whereas others are more suitable thanks to their high generalization capability,
such as support vector machine (SVM) [20] and artificial neural networks especially
those with many layers like convolutional neural networks (CNNs) which have
reached impressive performances in detection of AB from videos [21].
Semi-supervised detection.
In real-world applications, the supervised detection workflow described above is
not applicable due to the assumption of fully labelled data, on the basis of which ab-
normalities are known in advance and correctly labeled. However, dealing with elder-
Copyright © 2020 for this paper by its authors. Use permitted under Creative Com-
mons License Attribution 4.0 International (CC BY 4.0).
9
ly monitoring, abnormalities are not known in advance and obviously cannot be pur-
posely performed to train the detection algorithms. Semi-supervised detection is usu-
ally achieved by introducing the concept of one-class classification, whose state-of-
the-art implementations—as experimented in this study—are one-class SVM (OC-
SVM) [22] and auto-encoders (AEs) [23] within ML and DL fields, respectively. DL
techniques learn features in a hierarchical way: high-level features are derived from
low-level ones by using layer-wise pre-training, in such a way structures of every
higher level are represented in higher layers of the network. After pre-training, a
semi-supervised training provides a fine-tuning adjustment of the network via gradi-
ent descent optimization. Thanks to that greedy layer-wise pre-training followed by
semi-supervised fine-tuning [24], features can be automatically learned from large
datasets containing only one-class label, associated with normal behavior patterns.
Unsupervised Detection.
The most flexible workflow is that of unsupervised detection. It does not require
that abnormalities are known in advance but, conversely, they can occur during the
testing phase and are modelled as novelties with respect to normal (usual) observa-
tions. The main idea is that extracted features are scored solely on the basis of their
intrinsic properties. In order to decide what is normal and not, unsupervised detection
is based on appropriate metrics of either distance or density. The distance used in this
study is defined as follows:
𝐷(𝑠̅) = ∑𝑁
𝑖=1|𝑠𝑖 | ln(𝑖), (3)
where 𝑁 = 288 and 𝑠̅ = (𝑠1 , 𝑠2 , … , 𝑠𝑁 ) ∈ ℝ𝑁 is a day of observation, i.e., a row of
the matrix dataset.
Clustering techniques can be applied in unsupervised detection. In particular, K-
means is one of the simpler unsupervised algorithms that address the clustering prob-
lem by grouping data based on their similar features into K disjoint clusters. However,
K-means is affected by some shortcomings: (1) sensitivity to noise and outliers; (2)
initial cluster centroids (seeds) are unknown (randomly selected); and (3) there is no
criterion for determining the number of clusters. The Weighted K-Means [25], also
adopted in this study, provides a viable way to approach clustering of noisy data. The
last two problems are addressed by implementing the intelligent K-means suggested
by [26], in which the K-means algorithm is initialized by using the so-called anoma-
lous clusters, extracted before running the K-means itself.
3.3 Experimental setting
For the experimental purpose, 31500 datasets were generated, i.e., 1500 random in-
stances for each of the 21 abnormalities, obtained by considering the abnormalities
(St, Du, Di, Sw, Lo, Hr), together with all pairs of these abnormalities taken without
repetitions. Each dataset represented a 1-year data collection, as a matrix (image) of
365 rows (days) and 288 columns (samples lasting 5 min each), for a total amount of
105120 values (pixels) through 120 levels. The feature extraction process was carried
Copyright © 2020 for this paper by its authors. Use permitted under Creative Com-
mons License Attribution 4.0 International (CC BY 4.0).
10
out by considering a 50%-overlapping sliding window lasting 25 days, then leading to
a feature space of dimension D = 7200.
Each dataset was divided into three parts: Upper (1st–90th days), middle (90th–
180th days), and lower (180th–365th days) regions. The feature vectors (i.e., ACT-
LOC-HRL sequences) belonging to the upper regions were negative samples (i.e.,
normal behavior), whereas those belonging to the lower regions were positive ones
(i.e., AB). The middle regions were, instead, considered as prediction regions, charac-
terized by gradual changes becoming progressively more accentuated. The aim is to
classify the feature vectors belonging to the middle regions in order to predict the
likelihood of a future change which will become increasingly relevant and stable from
the 180th day onwards (lower region).
In both supervised and semi-supervised settings, regarding the SVM classifier, a
radial basis function (RBF) kernel was used. The kernel scale was automatically se-
lected using a grid search combined with cross-validation on randomly subsampled
training data. Regarding the CNN-based supervised detection, the network structure
included eight layers: four convolutional layers with a kernel size of 3 × 3, two sub-
sampling layers, and two fully connected layers. Finally, the two output units repre-
sented, via binary logical regression, the probability of normal and abnormal pattern
behaviors.
The stacked auto-encoder (SAE) network was structured in four hidden layers, and
the sliding-window feature vectors were given as input to the first layer, which thus
included 7200 units. The second hidden layer was of 900 units, corresponding to a
compression factor of 8 times. The following two hidden layers were of 180 and 60
units, respectively, with compression factors of 5 and 3 times. In supervised detection
settings, the six abnormal datasets were joined in order to perform a 6-fold cross-
validation scheme. In semi-supervised detection settings, instead, only normal data
from the same dataset were used for training, while testing was carried out using data
from day 90 onwards.
Regarding the convolutional auto-encoder (CAE) structure, the encoder included
three convolutional layers with a kernel size of five, five, and three, respectively,
followed by a fully connected layer. The decoder structure was a mirror of the encod-
er one. All experiments were performed on an Intel i7 3.5 GHz workstation with
16GB DDR3 and equipped with GPU NVidia Titan X using Keras [27] with Theano
[28] toolkit for DL approaches, and Matlab [29] for ML approaches.
4 Results and discussion
The achieved results are reported in Table 2 in terms of specificity, sensitivity and
LTP for all learning techniques evaluated in the present study. Note that the latter is
related to SEN and SPE since LTP refers to the average number of days, before the
180th day (from which the changed behavior becomes stable), at which the change
can be detected with such sensitivity and specificity. The longer the LTP the earlier
the change can be predicted.
Copyright © 2020 for this paper by its authors. Use permitted under Creative Com-
mons License Attribution 4.0 International (CC BY 4.0).
11
Generally, detecting ABs by using supervised methods presents the shortcoming of
requiring both positive samples (i.e., changing activity sequences) and negative sam-
ples (i.e., habitual activity sequence) for model training. The SVM-based and CNN-
based detection have been evaluated by training models with positive and negative
samples taken from different datasets, in order to reproduce more accurately the real-
life conditions mentioned above. Due to the lack of real data for training discussed
above, the supervised approaches achieved the lowest detection performances.
The problem of training data availability is mitigated with semi-supervised tech-
niques, since only negative samples (i.e., normal behaviors) are required, which are
quite abundant in everyday activities. However, the main difficulty is to select train-
ing samples that are most representative of normal behaviors. The semi-supervised
approaches evaluated in this study, i.e., OC-SVM and SAE, achieved intermediary
detection performances, although with lower prediction performance (LTP) due to the
difficulty selection of suitable (negative) samples for training.
The most promising results were obtained with the unsupervised learning methods,
i.e., K-means (KM) and DC, in which no labeled data were necessary, allowing the
easy adaptability to different environmental conditions as well as to users’ physical
characteristic and habits [13]. The KM detection, however, required an initial obser-
vation period during which the system was unable to detect changes from usual activi-
ty, negatively affecting the resulting prediction performance.
Classical ML methods, such as SVM and OC-SVM, have to deal with the problem
of learning a probability distribution from a set of samples, which generally means to
learn a probability density that maximizes the likelihood on given data. Conversely,
such density does not always exist, as what happens when data lie on low-
dimensional manifolds, which is the case of change types involving a narrow range of
values (e.g., if the change regards only few heartrate levels), or when training and
testing data come from a different probability distribution (e.g., as in the case of su-
pervised learning). Under such a point of view, conversely, DL methods are more
effective because they follow an alternative approach. Instead of attempting to esti-
mate a density, which may not exist, they define a parametric function (deep network)
able to generate samples closer to data samples taken from the original data distribu-
tion (by hyper-parameter tuning).
Table 2. Detection performance and lead-time of prediction for each technique.
SVM CNN OCSVM SAE KM CAE
SEN (%) 87 90 89 93 96 98
SPE (%) 91 92 90 92 97 99
LTP (days) 10 11 6 9 14 22
5 Conclusions
The contribution of this study is twofold. First, a common data model able to rep-
resent and process simultaneously ADLs, home locations and vital signs as image
Copyright © 2020 for this paper by its authors. Use permitted under Creative Com-
mons License Attribution 4.0 International (CC BY 4.0).
12
matrices is presented. Second, the performance of state-of-the-art ML-based and DL-
based detection techniques have been evaluated by considering large datasets, syn-
thetically generated, including both normal and abnormal behaviors. The achieved
results are promising and show the superiority of DL-based techniques in dealing with
huge datasets characterized by different kinds of data distribution. Future and ongoing
activities are focused on the evaluation of ML/DL learning techniques in different
domains, such as clinical decision support system and predictive maintenance.
References
1. Sharma, R.; Nah, F.; Sharma, K.; Katta, T.; Pang, N.; Yong, A. Smart living for elderly:
De-sign and human-computer interaction considerations. Lect. Notes Comput. Sci. 2016,
9755, 112–122.
2. Parisa, R.; Mihailidis, A. A survey on ambient-assisted living tools for older adults. IEEE
J. Biomed. Health Inform. 2013, 17, 579–590.
3. Mabrouk, A.B.; Zagrouba, E. Abnormal behavior recognition for intelligent video surveil-
lance systems: A review. Expert Syst. Appl. 2018, 91, 480–491.
4. Bakar, U.; Ghayvat, H.; Hasanm, S.F.; Mukhopadhyay, S.C. Activity and anomaly detec-
tion in smart home: A survey. Next Gener. Sens. Syst. 2015, 16, 191–220.
5. Taraldsen, K.; Chastin, S.F.M.; Riphagen, I.I.; Vereijken, B.; Helbostad, J.L. Physical ac-
tivity monitoring by use of accelerometer-based body-worn sensors in older adults: A sys-
tematic literature review of current knowledge and applications. Maturitas 2012, 71, 13–
19.
6. Min, C.; Kang, S.; Yoo, C.; Cha, J.; Choi, S.; Oh, Y.; Song, J. Exploring current practices
for battery use and management of smartwatches. In Proceedings of the ISWC '15 Pro-
ceedings of the 2015 ACM International Symposium on Wearable Computers, Osaka, Ja-
pan, 07–11 September 2015.
7. Stara, V.; Zancanaro, M.; Di Rosa, M.; Rossi, L.; Pinnelli, S. Understanding the Interest
Toward Smart Home Technology: The Role of Utilitaristic Perspective. In ForItAAL;
Springer: Berlin, Germany, 2018.
8. Miao, Y.; Song, J. Abnormal Event Detection Based on SVM in Video Surveillance. In
Proceedings of the IEEE Workshop on Advance Research and Technology in Industry
Applications, Ottawa, ON, Canada, 29–30 September 2014.
9. Forkan, A.R.M.; Khalil, I.; Tari, Z.; Foufou, S.; Bouras, A. A context-aware approach for
long-term behavioural change detection and abnormality prediction in ambient assisted liv-
ing. Pattern Recognit. 2015, 48, 628–641.
10. Krizhevsky, A.; Sutskever, I.; Hinton, G.E. Imagenet classification with deep convolution-
al neural networks. In Proceedings of the Advances in neural information processing sys-
tems, Lake Tahoe, Nevada, 3–6 December 2012.
11. Hejazi, M.; Singh, Y.P. One-class support vector machines approach to anomaly detection.
Appl. Artif. Intell. 2013, 27, 351–366.
12. Krizhevsky, A.; Hinton, G.E. Using very deep autoencoders for content-based image re-
trieval. In Proceedings of the 19th European Symposium on Artificial Neural Networks,
Bruges, Belgium, 27–29 April 2011.
13. Otte, F.J.P.; Saurer, R.B.; Stork, W. Unsupervised Learning in Ambient Assisted Living
for Pattern and Anomaly Detection: A Survey. CCIS 2013, 413, 44–53.
14. De Amorim, R.C.; Mirkin, B. Minkowski metric, feature weighting and anomalous cluster
initializing in K-Means clustering. Pattern Recognit. 2012, 45, 1061–1075.
Copyright © 2020 for this paper by its authors. Use permitted under Creative Com-
mons License Attribution 4.0 International (CC BY 4.0).
13
15. Chiang, M.M.T.; Mirkin, B. Intelligent choice of the number of clusters in k-means clus-
tering: An experimental study with different cluster spreads. J. Classif. 2010, 27, 3–40.
16. Forkan, A.R.M.; Khalil, I.; Tari, Z.; Foufou, S.; Bouras, A. A context-aware approach for
long-term behavioural change detection and abnormality prediction in ambient assisted liv-
ing. Pattern Recognit. 2015, 48, 628–641.
17. Kim, E.; Helal, S.; Cook, D. Human activity recognition and pattern discovery. IEEE Per-
vasive Comput. 2010, 9.1, 48.
18. Antle, M.C.; Silver, R. Circadian insights into motivated behavior. In Behavioral Neuro-
science of Motivation; Springer: Berlin, Germany, 2015; pp. 137–169.
19. Caroppo, A.; Diraco, G.; Rescio, G.; Leone, A.; Siciliano, P. Heterogeneous sensor plat-
form for circadian rhythm analysis. In Proceedings of the 6th IEEE International Work-
shop on Advances in Sensors and Interfaces (IWASI), Gallipoli, Italy, 10 August 2015,
pp. 187-192.
20. Hu, W.; Liao, Y.; Vemuri, V.R. Robust anomaly detection using support vector machines.
In Proceedings of the international conference on machine learning, Los Angeles, Califor-
nia (USA), 23-24 June 2003; pp. 282–289.
21. Sabokrou, M.; Fayyaz, M.; Fathy, M.; Moayed, Z.; Klette, R. Deep-anomaly: Fully convo-
lutional neural network for fast anomaly detection in crowded scenes. Comput. Vis. Image
Underst. 2018, 172, 88–97.
22. Hejazi, M.; Singh, Y.P. One-class support vector machines approach to anomaly detection.
Appl. Artif. Intell. 2013, 27, 351–366.
23. Krizhevsky, A.; Hinton, G.E. Using very deep autoencoders for content-based image re-
trieval. In Proceedings of the 19th European Symposium on Artificial Neural Networks,
Bruges, Belgium, 27–29 April 2011.
24. Erhan, D.; Bengio, Y.; Courville, A.; Manzagol, P.A.; Vincent, P.; Bengio, S. Why does
unsupervised pre-training help deep learning? J. Mach. Learn. Res. 2010, 11, 625–660.
25. De Amorim, R.C.; Mirkin, B. Minkowski metric, feature weighting and anomalous cluster
initializing in K-Means clustering. Pattern Recognit. 2012, 45, 1061–1075.
26. Chiang, M.M.T.; Mirkin, B. Intelligent choice of the number of clusters in k-means clus-
tering: An experimental study with different cluster spreads. J. Classif. 2010, 27, 3–40.
27. Chollet, F. Keras. GitHub repository. Available online: https://github.com/fchollet/keras
(accessed on February 12, 2019).
28. Bastien, F.; Lamblin, P.; Pascanu, R.; Bergstra, J.; Goodfellow, I.J.; Bergeron, A.; Bou-
chard, N.; Bengio, Y. Theano: New Features and Speed Improvements. Deep Learning and
Unsupervised Feature Learning NIPS Workshop: Lake Tahoe, California/Nevada (USA)
2012.
29. Matlab R2014; The MathWorks, Inc.: Natick, MA, USA. Available online:
https://it.mathworks.com (accessed on March 21, 2014).
Copyright © 2020 for this paper by its authors. Use permitted under Creative Com-
mons License Attribution 4.0 International (CC BY 4.0).