Framework for the Recognition of Activities of Daily Living and

Their Environments in the Development of a

Personal Digital Life Coach

Ivan Miguel Pires

1,2,3

, Nuno M. Garcia

, Nuno Pombo

and Francisco Flórez-Revuelta

Instituto de Telecomunicações, Universidade da Beira Interior, Covilhã, Portugal

Altranportugal, Lisbon, Portugal

ALLab - Assisted Living Computing and Telecommunications Laboratory, Computer Science Department,

Universidade da Beira Interior, Covilhã, Portugal

Department of Computer Technology, Universidad de Alicante, Spain

Keywords: Activities of Daily Living, Sensors, Mobile Devices, Data Fusion, Feature Extraction, Pattern Recognition.

Abstract: Due to the commodity of the use of the off-the-shelf mobile devices and technological devices by ageing

people, the automatic recognition of the Activities of Daily Living (ADL) and their environments using these

devices is a research topic were studied in the last years, but this project consists in the creation of an automatic

method that recognizes a defined dataset of ADL using a large set of sensors available in these devices, such

as the accelerometer, the gyroscope, the magnetometer, the microphone and the Global Positioning System

(GPS) receiver. The fusion of the data acquired from the selected sensors allows the recognition of an

increasing number of ADL and environments, where the ADL are mainly recognized with motion, magnetic

and location sensors, but the environments are mainly recognized with acoustic sensors. During this project,

several methods have been researched in the literature, implementing three types of neural networks, these

are Multilayer Perceptron (MLP) with Backpropagation, Feedforward neural network (FNN) with

Backpropagation and Deep Neural Networks (DNN), verifying that the neural networks that report highest

results are the DNN method for the recognition of ADL and standing activities, and the FNN method for the

recognition of environments.

1 INTRODUCTION

Mobile devices has several sensors embedded that are

capable for the acquisition of physical and

physiological parameters for the recognition of

Activities of Daily Living (ADL) and their

environments. The sensors commonly available in the

off-the-shelf mobile devices are the accelerometer,

the gyroscope, the magnetometer, the microphone,

and the Global Positioning System (GPS) receiver.

The use of these sensors in a system for the

monitoring of the lifestyle and/or the elderly people,

and the training of the lifestyles is included in the

research about the Ambient Assisted Living (AAL)

systems.

These sensors are available in the equipments

used daily, but their capabilities are not widely

explored, and this paper presents the development of

a new framework for the recognition of ADL and

their environments (Pires et al., 2016-a; Pires et al.,

2015; Pires et al., 2016-b), taking in account the

limitations of these devices, but achieving reliable

results for further implementation in the development

of a personal digital life coach (Garcia, 2016). As

presented in the figure 1, this framework has several

stages, such as data acquisition, data processing, data

fusion, and classification methods. This project is

already started and some results were achieved,

exploring the use of several types of neural networks

in the recognition of the ADL and their environments,

these are the Multilayer Perceptron (MLP) with

Backpropagation, the Feedforward neural network

(FNN) with Backpropagation, and the Deep Neural

Networks (DNN). The currently achieved results are

available in (Pires et al., 2017 (In Review)-a; Pires et

al., 2017 (In Review)-b; Pires et al., 2017 (In

Review)-c; Pires et al., 2017 (In Review)-d) and the

data acquired for the experiments are available in a

free repository (ALLab, 2017).

Pires, I., Garcia, N., Pombo, N. and Flórez-Revuelta, F.

Framework for the Recognition of Activities of Daily Living and Their Environments in the Development of a Personal Digital Life Coach.

DOI: 10.5220/0006824301630170

In Proceedings of the 7th International Conference on Data Science, Technology and Applications (DATA 2018), pages 163-170

ISBN: 978-989-758-318-6

163

Figure 1: Workflow of the proposed framework for the recognition of ADL and their environments.

2 RELATED WORK

2.1 Data Acquisition and Processing

2.1.1 Data Acquisition

Data acquisition process using mobile devise is

commonly performed without the use of frameworks,

but there are some studies using frameworks, e.g.,

Acquisition Cost-Aware QUery Adaptation

(ACQUA) that performs dynamic modification in the

order of the data acquisition and the streams

requested from the different sensors (Lim et al.,

2012). However, in the major part of the studies, the

data acquisition does not use a framework for the data

acquisition, reading directly the data from each sensor

available (Scalvini et al., 2013).

2.1.2 Data Cleaning

The data cleaning is the process to filter the data

acquired from the sensors in order to remove or fix

the incorrect values commonly named as noise, using

different types of filters based on the type of data

acquired (Jeffery et al., 2006). Firstly, for the data

acquired from the accelerometer, gyroscope and

magnetometer sensors, the filter that is commonly

applied are the low pass filter (Graizer, 2012).

Finally, for the data acquired from the microphone,

the filter that is commonly applied is the Fast Fourier

Transform (FFT) (Rader and Brenner, 1976) for the

extraction of the frequencies.

2.1.3 Data Imputation

During the data acquisition, several factors may cause

the loss of the data, the hardware fails, the positioning

of the mobile device, the different sampling rate

between the several sensors used, and the number of

sensors used (Bersch et al., 2014). Our previous study

(Pires et al., 2016-a) presents several methods for the

performance of the validation of the data acquired,

that may have different types, these are Missing

Completely At Random (MCAR), Missing At

Random (MAR) and Missing Not At Random

(MNAR) (Vateekul and Sarinnapakorn, 2009).

Based on the literature, the most used method for

the imputation of the sensors’ data is the K-Nearest

Neighbor (k-NN) and their variants (García-Laencina

et al., 2009), but there are other methods used for data

imputation, these are mean imputation (MEI)

(Rahman et al., 2015), multiple imputation (Ni et al.,

2005), linear regression (D’Ambrosio et al., 2012),

logistic regression (D’Ambrosio et al., 2012), among

others.

2.1.4 Feature Extraction

There are several studies that uses different features,

based on the purpose of the study and the sensors

used, but the correct definition of the features is

important to improve the accuracy of the methods for

the recognition of the different ADL.

Related to the extraction of the features from the

accelerometer, gyroscope and magnetometer sensors,

the most extracted features are the mean (Liu et al.,

2016), the variance (Liu et al., 2016), the maximum

DATA 2018 - 7th International Conference on Data Science, Technology and Applications

164

(Liu et al., 2016), the minimum (Liu et al., 2016), the

standard deviation (Liu et al., 2016), the average time

between peaks (Kumar and Gupta, 2015), among

others.

Related to the extraction of the features from the

microphone data, the most extracted features are the

average value (Hon et al., 2015), the threshold value

(Hon et al., 2015), the minimum value (Hon et al.,

2015), the maximum value (Hon et al., 2015), and

Mel-frequency cepstrum coefficients (MFCC) (Sert

et al., 2006)

Related to the extraction of the features from the

GPS receiver, the most extracted features are the

distance travelled (Shoaib et al., 2013), the speed

(Shoaib et al., 2013), and the location (Shoaib et al.,

2013; Zou et al., 2016).

2.2 Data Fusion and Classification

2.2.1 Recognition of Common ADL

After the extraction of several features from the

accelerometer, magnetometer and gyroscope sensors,

they need to be fused for the application of

classification methods for the recognition of ADL.

The authors of (Guo et al., 2016) recognized the

sitting, standing, walking, walking on stairs, and

running activities, using the accelerometer,

gyroscope and magnetometer sensors’ data and

applying the Random Forest classifier with several

features, such as the variance, the mean, the

frequency of the point with maximum amplitude, the

energy of the extremum value, the mean of the

extremum value, the sum of the difference between

extremum values, among others.

The authors of (Shoaib et al., 2013) used several

methods, including Artificial Neural Networks

(ANN), Support Vector Machine (SVM), Naïve

Bayes, Logistic regression, decision tree, k-NN, and

rule based classifiers, for the recognition of sitting,

standing, walking, walking on stairs, and running

activities, using the mean and the standard deviation

as features extracted from the accelerometer,

magnetometer and gyroscope sensors.

In (Elhoushi, Georgy, Wahdan, Korenberg, and

Noureldin, 2014), several features were extracted

from the accelerometer, magnetometer, and

gyroscope sensors, including the mean, the median,

the variance, the standard deviation, the inter-quartile

range, the Zero-Crossing Rate and the number of

peaks, and they implemented a decision tree method

for the recognition of walking on stairs, walking on

an escalator, standing and taking an elevator.

2.2.2 Recognition of Environments

After the extraction of several features from the

microphone and other sensors, they need to be fused

for the application of classification methods for the

recognition of environments and other ADL. The

authors of (Lane et al., 2011) extracted the spectral

roll-off from the microphone data and other features

extracted from the other sources used for the correct

recognition of walking, sleeping, running, standing,

and social interaction activities, based on the

environment, using linear and logistic regression

methods.

The authors of (Mengistu et al., 2016) used the

Support Vector Machine (SVM) and Gradient

Boosting Decision Tree methods with the zero-

crossing rate, the total spectrum power, the sub-band

powers, the spectral centroid, the spectral spread, the

spectral flux, the spectral roll-off, and the Mel-

Frequency Cepstral Coefficients (MFCC) as features

extracted from the microphone and other features

extracted from other sources for the recognition of

standing, lying, walking, walking on stairs, jogging,

drinking and running activities based on the

environment.

In (Nishida et al., 2015), the Gaussian mixture

model (GMM) was used with the log power and the

MFCC as features extracted from the microphone for

the recognition of cycling, cleaning table, shopping,

travelling by car, going to toilet, cooking, watching

television, eating, driving, working on a computer,

reading, and sleeping activities based on the

environment.

The accelerometer and the microphone was used

by the authors of (Filios et al., 2015) for the

recognition of several activities based on the

environment, including shopping, waiting in a queue,

driving, travelling by car, cleaning with a vacuum

cleaner, cooking, washing dishes, working at a

computer, sleeping, watching television, being a bar,

sitting, walking, standing, lying, and standing,

extraction the mean, the standard deviation, the range,

the angular degree, and the MFCC as features for the

application of decision tree methods and the IBk lazy

algorithm.

2.2.3 Recognition of Standing Activities

After the extraction of several features from the GPS

receiver and other sensors, they need to be fused for

the application of classification methods for the

recognition of ADL. The authors of (Shoaib et al.,

2013) used several methods, including ANN, SVM,

Naïve Bayes, Logistic regression, decision tree, k-

NN, and rule based classifiers, for the recognition of

Framework for the Recognition of Activities of Daily Living and Their Environments in the Development of a Personal Digital Life Coach

165

sitting, standing, walking, walking on stairs, and

running activities, using the distance, the location and

the speed as features extracted from the GPS receiver,

and other features extracted from other sensors.

The distance the location and the speed are also

extracted from the GPS receiver by the authors of

(Hung et al., 2014), and other features were extracted

from other sensors in order to recognize walking,

standing, walking on stairs, lying and running

activities, using J48 decision tree, Logistic

Regression, ANN, and SVM methods.

In (Altini et al., 2014), the SVM method was

implemented with the altitude difference in meters

and speed extracted as features from the GPS receiver

and other features extracted from other sources, in

order to recognize sitting, standing, washing dishes,

walking on stairs, cycling, and running.

The authors of (Luštrek et al., 2015) extracted the

distance between to access points was inputted as

feature from the GPS receiver and other features were

extracted from other sources, using the Naïve Bayes,

C4.5 decision tree, RIPPER, SVM, Random Forest,

Bagging, AdaBoost and Vote methods for the

recognition of sleeping, standing, preparing food,

eating, working, jogging, and travelling.

3 METHODS

3.1 Data Acquisition and Processing

3.1.1 Data Acquisition

This step includes the development of a mobile

application that acquires the data from several sources

available in the Android devices, these are

accelerometer, gyroscope, magnetometer,

microphone and GPS receiver. The data was acquired

in a background process and in real life environment

with the mobile device in the pocket for the

recognition of the signal of the sensors. The

population included in the experiments is aged

between 16 and 60 years old, performing several

activities and providing their feedback with the

selection of the activity performed. The ADL

included in this study are sleeping, walking on stairs,

walking, running, standing and driving. In addition,

the environments recognized are bar, gym, kitchen,

classroom, library, hall, street, bedroom, and

watching TV. The data acquired for this project is

available in a free repository (ALLab, 2017).

3.1.2 Data Cleaning

The application of the data cleaning methods depends

on the type of sensors used during the data acquisition

method presented in the section 3.1.1. When the study

is based on data acquired from the motion and

magnetic sensors, e.g., accelerometer, gyroscope and

magnetometer sensors, the best method for the data

cleaning process is the low pass filter (Graizer, 2012).

However, when the study makes use of acoustic data,

the best method for the data cleaning is based on the

extraction of the relevant frequencies using the Fast

Fourier Transform (FFT) (Rader and Brenner, 1976).

Related to the location sensors’ data, the data cleaning

methods are not useful for the improvement of the

recognition of ADL and their environments.

3.1.3 Data Imputation

The imputation of the sensors’ data acquired with the

mobile application may improve the reliability of the

framework for the recognition of ADL and their

environments. There are some problems that can be

minimized with data imputation methods, where the

most used methods are the mean imputation (MEI)

(Rahman et al., 2015), and the K-Nearest Neighbor

(k-NN) (García-Laencina et al., 2009).

3.1.4 Feature Extraction

After the application of the methods presented in the

previous sections and based on the sensors used in the

framework for the recognition of ADL and their

environments, we are able to extract the different

features, these are:

 Accelerometer, gyroscope and magnetometer

sensors’ data (stage 1): 5 greatest distances

between the maximum peaks, average, standard

deviation, variance and median of the maximum

peaks, standard deviation, average, variance,

maximum, minimum and median of the raw

signal;

 Microphone data (stage 2): 26 MFCC

coefficients, standard deviation, average,

maximum, minimum, variance and median of the

raw signal;

 Accelerometer, gyroscope and magnetometer

sensors’ data, microphone data and GPS receiver

data (stage 3): 5 greatest distances between the

maximum peaks, average, standard deviation,

variance and median of the maximum peaks,

standard deviation, average, variance, maximum,

minimum and median of the raw signal for the

accelerometer, gyroscope and magnetometer

DATA 2018 - 7th International Conference on Data Science, Technology and Applications

166

sensors, the environment recognized, and the

distance and location extracted from the GPS

receiver.

3.2 Data Fusion and Classification

The proposed study includes the use of different types

of neural networks in order to identify the best

methods for each stage of the implementation of the

framework for the recognition of ADL and their

environments. The types of neural networks selected

for the experiments are:

 MLP method, applied with Neuroph framework

(Neuroph, 2017);

 FNN method, applied with Encog framework

(Research, 2017);

 DNN method, applied with DeepLearning4j

framework (Nicholson, 2017).

Table 1 summarizes the configurations of the neural

networks studied for the development of the

framework for the recognition of ADL and their

environments, which all of the neural networks

implemented use the Sigmoid as activation function

and backpropagation.

Table 1: Configurations of the classification methods.

Parameters MLP FNN DNN

Activation function Sigmoid Sigmoid Sigmoid

Learning rate 0.6 0.6 0.1

Momentum 0.4 0.4 N/A

Maximum number

of training iterations

4 x 10

Number of layers 1 1 3

Weight function N/A N/A Xavie

Seed value N/A N/A 6

Backpropagation Yes Yes Yes

Regularization N/A N/A L

3.2.1 Recognition of Common ADL

This stage includes the use of accelerometer,

magnetometer and gyroscope sensors for the

recognition of the most common ADL, these are

walking on stairs, running, walking and standing. For

this research, we implemented the three types of

neural networks presented in the section 3.2 with

normalized and non-normalized data as well as with

different sets of features. The normalization of the

data depends on the type of neural network

implements, and, for the implementation of the MLP

and Feedforward networks with Backpropagation, the

normalization method used was the MIN/MAX

normalizer (Jain et al., 2005), and, for the

implementation of the DNN method, the

normalization with mean and standard deviation

(Brocca et al., 2010) and the application of the L

regularization (Ng, 2004) were performed.

3.2.2 Recognition of Environments

This stage includes the use of the microphone data for

the recognition of the environments, these are bar,

gym, kitchen, classroom, library, hall, street,

bedroom, and watching TV. For this research, we

implemented the three types of neural networks

presented in the section 3.2 with normalized and non-

normalized data as well as with different sets of

features. The normalization of the data depends on the

type of neural network implements, and, for the

implementation of the MLP and FNN methods, the

normalization method used was the MIN/MAX

normalizer (Jain et al., 2005), and, for the

implementation of the DNN method, the

normalization with mean and standard deviation

(Brocca et al., 2010) and the application of the L

regularization (Ng, 2004) were performed.

3.2.3 Recognition of Standing Activities

This stage includes the use of accelerometer,

magnetometer and gyroscope sensors’ data, the

environment recognized with the method

implemented in the section 3.2.2, and the distance and

location features extracted from the GPS receiver for

the recognition of standing activities, these are

sleeping, driving and watching TV. For this research,

we implemented the three types of neural networks

presented in the section 3.2 with normalized and non-

normalized data as well as with different sets of

features. The normalization of the data depends on the

type of neural network implements, and, for the

implementation of the MLP and FNN methods, the

normalization method used was the MIN/MAX

normalizer (Jain et al., 2005), and, for the

implementation of the DNN method, the

normalization with mean and standard deviation

(Brocca et al., 2010) and the application of the L

regularization (Ng, 2004) were performed.

4 RESULTS

4.1 Recognition of Common ADL

For the development of the method for the

recognition of the common ADL, the reported results

with the different number of sensors allowed and with

normalized and non-normalized data are presented in

the tables 2 and 3.

Framework for the Recognition of Activities of Daily Living and Their Environments in the Development of a Personal Digital Life Coach

167

Table 2: Classification accuracies with non-normalized data

for common ADL.

MLP FNN DNN

Accelerometer 34.76% 74.45% 80.35%

Accelerometer and

Magnetometer

35.15% 42.75% 70.43%

Accelerometer,

Magnetometer and

Gyroscope

38.32% 76.13% 74.47%

As verified the best method for the different

number of sensors is the DNN method with

normalized data, where the reported results are

highlighted in the table 3, and they are between

85.89% and 89.51%.

Table 3: Classification accuracies with normalized data for

common ADL.

MLP FNN DNN

Accelerometer 24.03% 37.07% 85.89%

Accelerometer and

Magnetometer

24.93% 64.94%

86.49%

Accelerometer,

Magnetometer and

Gyroscope

37.13% 29.54%

89.51%

4.2 Recognition of Environments

For the development of the method for the

recognition of the environments, the reported results

with normalized and non-normalized data are

presented in the table 4, verifying that the best results

are achieved with the FNN method with non-

normalized data, reporting an accuracy of 86.50%.

Table 4: Classification accuracies with acoustic data.

MLP FNN DNN

Non-normalized 12.86% 86.50% 48.11%

Normalized 19.43% 82.75% 4.74%

4.3 Recognition of Standing ADL

The results of the recognition of the standing ADL

depends on the correct recognition of the common

ADL as standing, because, based on the environment

recognized and/or the distance travelled, the standing

ADL are recognized with a reported accuracy of

100%, based on the results of the DNN method with

normalized data.

4.4 Overall Results

The development of the proposed framework for the

recognition of ADL and their environments explored

several scenarios, showing the accuracies reported by

the selected method for each scenarios, where a sce-

nario is a combination of sensors used. They are:

A. Use of the accelerometer;

B. Use of the accelerometer and the magnetometer;

C. Use of the accelerometer, the magnetometer and

the gyroscope;

D. Use of the microphone;

E. Use of the environment recognized and/or the

GPS receiver.

Table 5 shows the accuracies of the selected method,

presenting the accuracy of the proposed framework

that, based on the number of sensors available, is

between 90.80% and 92%. Finally, the average

accuracy of the framework is 91.27%.

Table 5: Classification accuracies of the proposed

framework.

Stages

A / D /

B / D /

C / D /

Average

accuracy

Common

ADL

85.89% 86.49% 89.51% 87.30%

Environments 86.50% 86.50% 86.50% 86.50%

Standing

activities

100.00% 100.00% 100.00% 100.00%

Average

accuracy

90.80% 91.00% 92.00% 91.27%

5 CONCLUSIONS

The recognition of ADL and their environments using

the commodity off-the-shelf mobile is a project that

allows the training and monitoring of lifestyles with

reliable accuracy and the reducing costs in the

monitoring of elderly people and/or the physical

training as a personal trainer.

Several research have been performed using small

sets of sensors, but the current state of this project

probes that the use of a major number of sensors

increases the number of ADL and environments

recognized and the accuracy of the recognition.

The current development of this project using the

data available in (ALLab, 2017) and neural networks

reports an accuracy around 91.27%, complaining the

several stages of the treatment and analysis of the

sensors’ data.

As future work, the implementation of data

imputation methods and other classification methods,

including Adaboost and Support Vector Machines

(SVM), reveals important to attempt to increase the

reliability of the framework, improving the quality of

the data acquired. The development of the method

should take in account the limitations of the mobile

DATA 2018 - 7th International Conference on Data Science, Technology and Applications

168

devices. However, the ongoing results proves that the

combination of the different types of neural networks

achieves reliable results in the recognition.

ACKNOWLEDGEMENTS

This work was supported by FCT project

UID/EEA/50008/2013 (Este trabalho foi suportado

pelo projecto FCT UID/EEA/50008/2013).

The authors would also like to acknowledge the

contribution of the COST Action IC1303 – AAPELE

– Architectures, Algorithms and Protocols for

Enhanced Living Environments.

REFERENCES

ALLab. (2017). August 2017- Multi-sensor data fusion in

mobile devices for the identification of activities of

daily living - ALLab Signals. Retrieved from

https://allab.di.ubi.pt/mediawiki/index.php/August_20

17-_Multi-sensor_data_fusion_in_mobile_devices_

for_the_identification_of_activities_of_daily_living.

Altini, M., Vullers, R., Van Hoof, C., van Dort, M., and

Amft, O. (2014). Self-Calibration of Walking Speed

Estimations Using Smartphone Sensors. 2014 Ieee

International Conference on Pervasive Computing and

Communications Workshops (Percom Workshops), 10-

18. doi:10.1109/PerComW.2014.6815158.

Bersch, S. D., Azzi, D., Khusainov, R., Achumba, I. E., and

Ries, J. (2014). Sensor data acquisition and processing

parameters for human activity classification. Sensors

(Basel), 14(3), 4239-4270. doi:10.3390/s140304239.

Brocca, L., Melone, F., Moramarco, T., Wagner, W.,

Naeimi, V., Bartalis, Z., and Hasenauer, S. (2010).

Improving runoff prediction through the assimilation of

the ASCAT soil moisture product. Hydrology and

Earth System Sciences, 14(10), 1881-1893.

doi:10.5194/hess-14-1881-2010.

Chris Nicholson, A. (2017). Deeplearning4j: Open-source,

Distributed Deep Learning for the JVM. Retrieved

from https://deeplearning4j.org/

D’Ambrosio, A., Aria, M., and Siciliano, R. (2012).

Accurate Tree-based Missing Data Imputation and Data

Fusion within the Statistical Learning Paradigm.

Journal of Classification, 29(2), 227-258.

doi:10.1007/s00357-012-9108-1.

Elhoushi, M., Georgy, J., Wahdan, A., Korenberg, M., and

Noureldin, A. (2014). Using Portable Device Sensors to

Recognize Height Changing Modes of Motion. 2014

Ieee International Instrumentation and Measurement

Technology Conference (I2mtc) Proceedings, 477-481.

doi:10.1109/i2mtc.2014.6860791.

Filios, G., Nikoletseas, S., Pavlopoulou, C., Rapti, M., and

Ziegler, S. (2015). Hierarchical Algorithm for Daily

Activity Recognition via Smartphone Sensors. 2015

IEEE 2nd World Forum on Internet of Things (Wf-Iot),

381-386. doi:10.1109/WF-IoT.2015.7389084.

Garcia, N. M. (2016). A Roadmap to the Design of a

Personal Digital Life Coach ICT Innovations 2015:

Springer.

García-Laencina, P. J., Sancho-Gómez, J.-L., Figueiras-

Vidal, A. R., and Verleysen, M. (2009). K nearest

neighbours with mutual information for simultaneous

classification and missing data imputation.

Neurocomputing, 72(7-9), 1483-1493. doi:10.1016/

j.neucom.2008.11.026.

Graizer, V. (2012). Effect of low‐pass filtering and re‐

sampling on spectral and peak ground acceleration in

strong‐motion records. Paper presented at the Proc.

15th World Conference of Earthquake Engineering,

Lisbon, Portugal.

Guo, Q., Liu, B., and Chen, C. W. (2016). A Two-layer and

Multi-strategy Framework for Human Activity

Recognition using smartphone. 2016 IEEE

International Conference on Communications (Icc), 1-

6. doi:10.1109/icc.2016.7511487.

Hon, T. K., Wang, L., Reiss, J. D., and Cavallaro, A. (2015).

Audio Fingerprinting for Multi-Device Self-

Localization. IEEE/ACM Transactions on Audio,

Speech, and Language Processing, 23(10), 1623-1636.

doi:10.1109/TASLP.2015.2442417.

Hung, W. C., Shen, F., Wu, Y. L., Hor, M. K., and Tang, C.

Y. (2014). Activity Recognition with Sensors on

Mobile Devices. Proceedings of 2014 International

Conference on Machine Learning and Cybernetics

(ICMLC), Vol 2, 449-454. doi:10.1109/icmlc.2014.

7009650.

Jain, A., Nandakumar, K., and Ross, A. (2005). Score

normalization in multimodal biometric systems.

Pattern recognition, 38(12), 2270-2285.

doi:10.1016/j.patcog.2005.01.012.

Jeffery, S. R., Alonso, G., Franklin, M. J., Hong, W., and

Widom, J. (2006). Declarative Support for Sensor Data

Cleaning. Paper presented at the Proceedings of

PERVASIVE'06 Proceedings of the 4th international

conference on Pervasive Computing, Dublin, Ireland.

Kumar, A., and Gupta, S. (2015). Human Activity

Recognition through Smartphone’s Tri-Axial

Accelerometer using Time Domain Wave Analysis and

Machine Learning. International Journal of Computer

Applications, 127(18), 22-26.

doi:10.5120/ijca2015906733.

Lane, N. D., Mohammod, M., Lin, M., Yang, X., Lu, H.,

Ali, S., Campbell, A. (2011). Bewell: A smartphone

application to monitor, model and promote wellbeing.

Paper presented at the 5th international ICST

Conference on Pervasive Computing Technologies for

Healthcare.

Lim, L., Misra, A., and Mo, T. (2012). Adaptive data

acquisition strategies for energy-efficient, smartphone-

based, continuous processing of sensor streams.

Distributed and Parallel Databases, 31(2), 321-351.

doi:10.1007/s10619-012-7093-3.

Liu, Y. Y., Zhao, F., Shao, W. H., and Luo, H. Y. (2016).

An Hidden Markov Model based Complex Walking

Framework for the Recognition of Activities of Daily Living and Their Environments in the Development of a Personal Digital Life Coach

169

Pattern Recognition Algorithm. Proceedings of 2016

Fourth International Conference on Ubiquitous

Positioning, Indoor Navigation and Location Based

Services (IEEE Upinlbs 2016), 223-229.

doi:10.1109/upinlbs.2016.7809976.

Luštrek, M., Cvetkovic, B., Mirchevska, V., Kafalı, Ö.,

Romero, A., and Stathis, K. (2015). Recognising

lifestyle activities of diabetic patients with a

smartphone. doi:10.4108/icst.pervasivehealth.

2015.259118.

Mengistu, Y., Pham, M., Do, H. M., and Sheng, W. (2016).

AutoHydrate: A Wearable Hydration Monitoring

System. 2016 IEEE/RSJ International Conference on

Intelligent Robots and Systems (Iros 2016), 1857-1862.

doi:10.1109/iros.2016.7759295.

Neuroph. (2017). Java Neural Network Framework

Neuroph. Retrieved from http://neuroph.sourceforge.

net/

Ng, A. Y. (2004). Feature selection, L 1 vs. L 2

regularization, and rotational invariance. Paper

presented at the Proceedings of the twenty-first

international conference on Machine learning.

Ni, D., Leonard, J. D., Guin, A., and Feng, C. (2005).

Multiple Imputation Scheme for Overcoming the

Missing Values and Variability Issues in ITS Data.

Journal of Transportation Engineering, 131(12), 931-

938. doi:10.1061/(asce)0733-947x(2005)131:12(931).

Nishida, M., Kitaoka, N., and Takeda, K. (2015). Daily

activity recognition based on acoustic signals and

acceleration signals estimated with Gaussian process.

Paper presented at the 2015 Asia-Pacific Signal and

Information Processing Association Annual Summit

and Conference (APSIPA).

Pires, I., Garcia, N., Pombo, N., and Flórez-Revuelta, F.

(2016). From Data Acquisition to Data Fusion: A

Comprehensive Review and a Roadmap for the

Identification of Activities of Daily Living Using

Mobile Devices. Sensors, 16(2), 184.

Pires, I. M., Garcia, N. M., and Flórez-Revuelta, F. (2015).

Multi-sensor data fusion techniques for the

identification of activities of daily living using mobile

devices. Paper presented at the Proceedings of the

ECMLPKDD 2015 Doctoral Consortium, European

Conference on Machine Learning and Principles and

Practice of Knowledge Discovery in Databases, Porto,

Portugal.

Pires, I. M., Garcia, N. M., Pombo, N., and Flórez-

Revuelta, F. (2016-a). Identification of Activities of

Daily Living Using Sensors Available in off-the-shelf

Mobile Devices: Research and Hypothesis. Paper

presented at the Ambient Intelligence-Software and

Applications–7th International Symposium on Ambient

Intelligence (ISAmI 2016).

Pires, I. M., Garcia, N. M., Pombo, N., and Flórez-

Revuelta, F. (2017 (In Review)-a). A Multiple Source

Framework for the Identification of Activities of Daily

Living Based on Mobile Device Data.

arXiv:1711.00104.

Pires, I. M., Garcia, N. M., Pombo, N., and Flórez-

Revuelta, F. (2017 (In Review)-b). User Environment

Detection with Acoustic Sensors Embedded on Mobile

Devices for the Recognition of Activities of Daily

Living. arXiv:1711.00124.

Pires, I. M., Garcia, N. M., Pombo, N., Flórez-Revuelta, F.,

and Rodríguez, N. D. (2016-b). Validation Techniques

for Sensor Data in Mobile Health Applications. Journal

of Sensors, 2016.

Pires, I. M., Garcia, N. M., Pombo, N., Flórez-Revuelta, F.,

and Spinsante, S. (2017 (In Review)-c). Data Fusion on

Motion and Magnetic Sensors embedded on Mobile

Devices for the Identification of Activities of Daily

Living. engrxiv.org/x4r5z.

Pires, I. M., Garcia, N. M., Pombo, N., Flórez-Revuelta, F.,

and Spinsante, S. (2017 (In Review)-d). Pattern

Recognition Techniques for the Identification of

Activities of Daily Living using Mobile Device

Accelerometer. arXiv:1711.00096.

Rader, C., and Brenner, N. (1976). A new principle for fast

Fourier transformation. IEEE Transactions on

Acoustics, Speech, and Signal Processing, 24(3), 264-

266. doi:10.1109/tassp.1976.1162805.

Rahman, S. A., Huang, Y., Claassen, J., Heintzman, N., and

Kleinberg, S. (2015). Combining Fourier and lagged k-

nearest neighbor imputation for biomedical time series

data. J Biomed Inform, 58, 198-207.

doi:10.1016/j.jbi.2015.10.004.

Research, H. (2017). Encog Machine Learning Framework.

Retrieved from http://www.heatonresearch.com/encog/

Scalvini, S., Baratti, D., Assoni, G., Zanardini, M., Comini,

L., and Bernocchi, P. (2013). Information and

communication technology in chronic diseases: a

patient’s opportunity. Journal of Medicine and the

Person, 12(3), 91-95. doi:10.1007/s12682-013-0154-1.

Sert, M., Baykal, B., and Yazici, A. (2006, 0-0 0). A Robust

and Time-Efficient Fingerprinting Model for Musical

Audio. Paper presented at the 2006 IEEE International

Symposium on Consumer Electronics.

Shoaib, M., Scholten, H., and Havinga, P. J. M. (2013).

Towards Physical Activity Recognition Using

Smartphone Sensors. 2013 IEEE 10th International

Conference on and 10th International Conference on

Autonomic and Trusted Computing (Uic/Atc)

Ubiquitous Intelligence and Computing, 80-87.

doi:10.1109/Uic-Atc.2013.43.

Vateekul, P., and Sarinnapakorn, K. (2009). Tree-Based

Approach to Missing Data Imputation. Paper presented

at the Data Mining Workshops, 2009. ICDMW '09.

IEEE International Conference on Miami, FL.

Zou, X., Gonzales, M., and Saeedi, S. (2016). A Context-

aware Recommendation System using smartphone

sensors. Paper presented at the 2016 IEEE 7th Annual

Information Technology, Electronics and Mobile

Communication Conference (IEMCON).

DATA 2018 - 7th International Conference on Data Science, Technology and Applications

170