More mobile & not so well-connected yet: users' mobility inference model and 6 month field study

TY - GEN

T1 - More mobile & not so well-connected yet

T2 - 7th International Congress on Ultra Modern Telecommunications and Control Systems and Workshops

AU - Wac, Katarzyna

AU - Pinar, Gerardo

AU - Gustarini, Mattia

AU - Marchanoff, Jerome

N1 - Conference code: 7

PY - 2016/1/13

Y1 - 2016/1/13

N2 - Smartphones assist their users throughout daily life activities. There is much emphasis on the user's mobility support in the research at large. However, we have a weak understanding about users mobility (are they really moving?) and how well connected are they across their typical day. First, to infer mobility state of users, we derived and evaluated the accuracy of a machine learning-based model, i.e., MobilitySensor, which is based solely on smartphone built-in sensors. It is a tree-based model, defined for each network operator and its average accuracy reaches 91%. Next, we leverage our algorithm to explore the mobility of 34 users served by 3 different Swiss operators (OP) during a period of six months, correlating it with their connectivity. The user study results showed that users are statistically significantly more mobile than we observed in the past (21±7% of the time, i.e., up to 4.3h vs. 13±12%, i.e., 2.7h in 2011) and when they are mobile, 4G network is available to them 38±12% of the time. Furthermore, when mobile, depending on their operator, they may be provided with up to around 10% of the time with 2.5G connectivity (for OP1 and OP2 vs. only 4% OP3), or provided mainly with 3G (49% for OP1 vs. 34% for OP3). Based on the results we provide a set of design implications for application providers, users and operators alike, all striving to improve the mobile users' quality of experience (QoE).

AB - Smartphones assist their users throughout daily life activities. There is much emphasis on the user's mobility support in the research at large. However, we have a weak understanding about users mobility (are they really moving?) and how well connected are they across their typical day. First, to infer mobility state of users, we derived and evaluated the accuracy of a machine learning-based model, i.e., MobilitySensor, which is based solely on smartphone built-in sensors. It is a tree-based model, defined for each network operator and its average accuracy reaches 91%. Next, we leverage our algorithm to explore the mobility of 34 users served by 3 different Swiss operators (OP) during a period of six months, correlating it with their connectivity. The user study results showed that users are statistically significantly more mobile than we observed in the past (21±7% of the time, i.e., up to 4.3h vs. 13±12%, i.e., 2.7h in 2011) and when they are mobile, 4G network is available to them 38±12% of the time. Furthermore, when mobile, depending on their operator, they may be provided with up to around 10% of the time with 2.5G connectivity (for OP1 and OP2 vs. only 4% OP3), or provided mainly with 3G (49% for OP1 vs. 34% for OP3). Based on the results we provide a set of design implications for application providers, users and operators alike, all striving to improve the mobile users' quality of experience (QoE).

KW - 3G mobile communication

KW - 4G mobile communication

KW - learning (artificial intelligence)

KW - mobile computing

KW - mobility management (mobile radio)

KW - quality of experience

KW - smart phones

KW - 2.5G connectivity

KW - 3G network

KW - 4G network

KW - OP

KW - QoE

KW - Swiss operator

KW - machine learning-based model

KW - mobility inference model

KW - smartphone

KW - tree-based model

KW - Context

KW - Current measurement

KW - Global Positioning System

KW - Mobile communication

KW - Mobile computing

KW - Quality of service

KW - Radiation detectors

KW - Mobility

KW - Quality of Experience

KW - Quality of Service

KW - Wireless communication

U2 - 10.1109/ICUMT.2015.7382411

DO - 10.1109/ICUMT.2015.7382411

M3 - Article in proceedings

SP - 91

EP - 99

BT - 2015 7th International Congress on Ultra Modern Telecommunications and Control Systems and Workshops (ICUMT)

PB - IEEE

Y2 - 6 October 2015 through 8 October 2015

ER -