Sociological Orbit based Mobility Profiling On Profiling Mobility

Sociological Orbit based Mobility Profiling On Profiling Mobility and Predicting Locations of Wireless Users Joy Ghosh Matthew J. Beal Hung Q. Ngo Chunming Qiao Lab for Advanced Network Design, Evaluation and Research

LANDER SOLAR: http: //www. cse. buffalo. edu/~joyghosh/solar. html Outline n n n n Introduction Mobility Traces Orbital Mobility Profiling Location Prediction Performance Results Comparison with contemporary work Other Applications of Mobility Profiling Future Directions Sociological Orbit based Mobility Profiling

LANDER SOLAR: http: //www. cse. buffalo. edu/~joyghosh/solar. html Mobile Users • influenced by social routines • visit a few “hubs” / places (outdoor/indoor) regularly • “orbit” around (fine to coarse grained) hubs at several levels Sociological Orbit based Mobility Profiling

LANDER SOLAR: http: //www. cse. buffalo. edu/~joyghosh/solar. html Hub Based Mobility Profiles and Prediction On any given day, a user may regularly visit a small number of “hubs” small number (e. g. , locations A andof B) “hubs” (e. g. , locations A and B) Each mobility is alistweighted listweight of hubs, n Each mobility profile isprofile a weighted of hubs, where = hub visit probability (e. g. , 70% A and 50% B) where weight =(e. g. , hub visit probability In any given period week), a user may n In any given period (e. g. , week), a user may follow a few such “mobility A “mobility and 50%profiles” B) profiles” (e. g. , P 1(e. g. , and P 2)70% follow a few such Eachis profile is in turn with(e. g. , a 60% P 1 n Each profile in turn associated withassociated a (daily) probability (e. g. , P 1 and P 2) and 40% P 2) (daily) P 1={A=0. 7, probability (e. g. , and 40%C=0. 6} P 2) Example: B=0. 5}60% and P 1 P 2={B=0. 9, n Example: P 1={A=0. 7, B=0. 5} and P 2={B=0. 9, C=0. 6} q On anan ordinary day, a user go to locations A, B C with the On ordinary day, may a user may go toand locations following probabilities, resp. : 0. 42 (=0. 6 x 0. 7), 0. 66 (= 0. 6 x 0. 5 + 0. 4+0. 9) and A, & C with the following probabilities: 0. 24 B(=0. 4 x 0. 6) q 0. 42 (=0. 6 x 0. 7), (= 0. 6 x 0. 5 + 0. 4+0. 9), (=0. 4 x 0. 6) 20% more accurate 0. 66 than simple visit-frequency based 0. 24 prediction q exactly which profile a user will follow on a given day prediction can result in • Knowing 20% more accurate than simple visit-frequency based more accurate prediction • even Knowing exactly which profile a user will follow on a given day can result in even more accurate prediction Sociological Orbit based Mobility Profiling

LANDER SOLAR: http: //www. cse. buffalo. edu/~joyghosh/solar. html Applications of Orbital Mobility Profiles n Location Predictions and Routing within MANET and ICMAN n Anomaly based intrusion detection unexpected movement (in time or space) sets off an alarm n Customizable traffic alerts alert only the individuals who might be affected by a specific traffic condition n Targeted inspection examine only the persons who have routinely visited specific regions n Environmental/health monitoring identify travelers who can relay data sensed at remote locations with no APs Sociological Orbit based Mobility Profiling

LANDER SOLAR: http: //www. cse. buffalo. edu/~joyghosh/solar. html Categories of Mobility Traces AP-based traces (system logs) collected by APs: MAC n. AP-based traces (system logs) collected by ID, AP ID, Time, AP Events n q Traces from Dartmouth, UNC APs: MAC ID, AP ETH ID, Zurich, Time, AP Events n. AP-based traces (system logs) by user devices n traces (system logs)UNC collected by q Traces from Dartmouth, ETH collected Zurich, (laptops, PDAs, etc. ) and uploaded periodically to central user devices (laptops, PDAs, etc. ) and server Traces from UCSD uploaded periodically to central server n. Pair wise user contact based traces without any n wisewireless based q. Traces from UCSD location information traces without location information HAGGLE project byany Intel Research n User location traces by user q. HAGGLE project bycollected Intelcollected Research n User location traces by user devices (based on GPS or other locationon tracking devices (based GPSdevices) or other location Recently available UMass. Diesel. Net traces devices) Thetracking last category is most appropriate for our study, but we started q q Recently available UMass. Diesel. Net traces our work with the first category, which was more accessible Sociological Orbit based Mobility Profiling

LANDER SOLAR: http: //www. cse. buffalo. edu/~joyghosh/solar. html Traces Used n Profiling techniques applied to ETH Zurich traces q q n Mapped APs into buildings based on AP’s coordinates, and each building becomes a “hub” q n Duration of 1 year from 4/1/04 till 3/31/05 13, 620 wireless users, 391 APs, 43 buildings Grouped users into 6 groups based on degree of activity Selected one sample (most active) user from each group Converted AP-based traces into hub-based traces Other traces q q Expect similar results from Dartmouth’s traces No sufficient AP location info from other traces UMass’s traces are for buses, more predictable than users Need to obtain actual users’ traces with GPS Sociological Orbit based Mobility Profiling

LANDER SOLAR: http: //www. cse. buffalo. edu/~joyghosh/solar. html Orbital Mobility Profiling n n n Obtain each user’s daily hub lists as binary vectors Represent each hub list (binary vector) as a point in a n-dimensional space (n = total number of hubs) Cluster these points into multiple clusters, each with a mean q q n n Using the Expectation-Maximization (EM) algorithm based on a Mixture of Bernoulli’s distribution Probe other classification methods: Bayesian-Bernoulli’s Each cluster mean represents a mobility profile, described as a probabilistic hub visitation list User’s mobility is aptly modeled using a mixture of mobility profiles with certain “mixing proportions” Sociological Orbit based Mobility Profiling

LANDER SOLAR: http: //www. cse. buffalo. edu/~joyghosh/solar. html Profiling illustration Translate to binary hub visitation vectors Obtain daily hub stay durations Sociological Orbit based Mobility Profiling Apply clustering algorithm to find mixture of profiles

LANDER SOLAR: http: //www. cse. buffalo. edu/~joyghosh/solar. html Profile parameters for all sample users Sociological Orbit based Mobility Profiling

LANDER SOLAR: http: //www. cse. buffalo. edu/~joyghosh/solar. html Hub-based Location Predictions - I n Unconditional Hub-visit Prediction q q Prediction Error = Incorrect hubs predicted over Total hubs SPE – Statistical based Prediction Error n n q PPE – Profile based Prediction Error n q SPE-ALL: (n+1)th day prediction based on hub-visit frequency from day 1 through day n SPE-W 7 : (n+1)th day prediction based on hub-visit frequency within last week, i. e. , day (n-7) through day n PPE-W 7 : (n+1)th day prediction based on profiles of the last week, i. e. , day (n-7) through day n Prediction Improvement Ration (PIR) n n PIR-ALL = (SPE-ALL – PPE-W 7) / SPE-ALL PIR-W 7 = (SPE-W 7 – PPE-W 7) / SPE-W 7 Sociological Orbit based Mobility Profiling

LANDER SOLAR: http: //www. cse. buffalo. edu/~joyghosh/solar. html Unconditional Prediction Results The profile mixing proportions vary with every window of n days Sociological Orbit based Mobility Profiling

LANDER SOLAR: http: //www. cse. buffalo. edu/~joyghosh/solar. html Hub-based Location Predictions - II n Conditional Hub-visit Prediction q Improvement given current profile is known/identifiable q It is possible sometimes to infer profile from current hub information alone q Our method effectively leverages information when available Actually visited Ht on day D or not Indicator Predicted probability Current Hub Profile based on profile Target Hub Predicted The ID: current will probability the day user in question visit using this visit hub? frequency Sample user categories (Current) Sociological Orbit based Mobility Profiling

LANDER SOLAR: http: //www. cse. buffalo. edu/~joyghosh/solar. html Hub-based Location Predictions - III n n n Hub sequence prediction based on hub transitional probability Prediction Accuracy = 1 – (incorrect predictions / total predictions) Scenario 1: only starting hub is known for sequence prediction Scenario 2: hub prediction is corrected at every hub in sequence Better performance with increasing knowledge – intuitive Statistical Profile Time based. Prediction. Accuracy(TPA) (PPA) (SPA)–––temporal no notime profile information profiles information Sociological Orbit based Mobility Profiling

LANDER SOLAR: http: //www. cse. buffalo. edu/~joyghosh/solar. html Comparison with contemporary work n Dartmouth – Infocom’ 06 q q n AP vs hub sequence of visit prediction Both built directed weighted graph based on AP/hub transition data AP/hub sequence prediction based on Markov Chain, Moving Average, CDF vs. Profile based in our case A comparison of these methods would be interesting Moby. Space – Infocom 2006 q q Each axis in the Moby. Space can be a location (instead of an AP) Each location is given a weight based on the frequency of visits Each user is represented by a point in Moby. Space, called Moby. Point, which is like the “cluster mean” (weighted hub list) in our case We have a mixture of profiles, enabling us to make better predictions Sociological Orbit based Mobility Profiling

LANDER SOLAR: http: //www. cse. buffalo. edu/~joyghosh/solar. html Comparison with contemporary work n Intel HAGGLE projects – Infocom’ 05 q q n May augment contact information with location information. But, every location where a pair come in contact (corridors, stairs) may have no significance later In our case, we give importance to social behavior and end up with fixed number of sociologically important hubs and only are interested in contacts within them. It is possible to determine contact probability based on hub movement with a Continuous Time Markov Chain (CTMC) that is provided with hub stay times (that follow a power law distribution) and the hub transitional probabilities If hubs are equipped with storage, two users sharing a hub in their profile can exchange messages UCL – MSWi. M ‘ 04 q q Actual mobility modeling based on social network theory Results from time series analysis will be useful for comparison with our profile based predictions Sociological Orbit based Mobility Profiling

LANDER SOLAR: http: //www. cse. buffalo. edu/~joyghosh/solar. html Future Directions n Work with other types of traces n Design other clustering/profiling techniques n Optimize techniques for mobility profile information dissemination, lifetime, maintenance and query n Compare location predictions with other methods based on e. g. , time series analysis n Study efficient routing algorithms Sociological Orbit based Mobility Profiling

Sociological Orbit based Mobility Profiling Thank You ! Questions? Lab for Advanced Network Design, Evaluation and Research