Smart Device Location Services:- A Reliable Analytics Resource? CORS/INFORMS, Montreal, June 2015 Richard J Self Senior Lecturer in Analytics and Governance University of Derby http://tinyurl.com/ppyg6t8 http://computing.derby.ac.uk email: r.j.self@derby.ac.uk Richard J Self - University of Derby 1 Based on Final Year Student Project 12 students researching 7 students contributed data to this analysis (2460 data points) Daniel Corah Vishal Patel Amna Almutawa Ishwa Khadka Victor Horecny Shehzaad kashmiri Farondeep Bains Richard J Self - University of Derby 2 Context (1) GPS accuracy claim: 95% of all fixes to be <=10m Thinknear identify the fact that 46% of reported locations are accurate <= 1000m (Q1 2015 Location Score report) 10% error > 100,000m (60 miles) My students’ research indicates (2420 data points) 85% are accurate to <= 25m 2.5% are >= 500m Outliers 1km to 80km Richard J Self - University of Derby 3 The Vs of Big Data and Analytics Big Data Veracity Over 80% of all data (small, large and big) is of uncertain veracity (J Easton, IBM, 2012, http://www.thebigdatainsightgroup.com/site/system/files/private_1 ) The critical Vs for A-GPS LS Veracity Variability Verification Visualisation Richard J Self - University of Derby 4 Critical Governance Questions What is the reliability of A-GPS in smart devices? What are the consequences of uncertain veracity of A-GPS based Location Services to relevant stakeholders? Richard J Self - University of Derby 5 Agenda Identify typical uses of LBS Evaluate accuracy of LBS in smart devices Identify governance issues of the use of LBS Richard J Self - University of Derby 6 Some Uses for LBS Marketing Recreational Social media Photo tagging European e-Call Geo-fencing? Car crash reporting (required max error of 100 – 200m) Crime prevention services GPS tagging Richard J Self - University of Derby 7 Triggers to Research Project wandering 22km error Nighttime wandering 4900m error from top of Mont-Royal Start-up movement V Patel – Key Insight – Models Vary phone N Mean Std Dev Std Err Nexus 54 41.5629 24.1146 3.2816 iPhone 58 85.5101 113.8 14.9403 Method Variances DF t Value Pr > |t| Pooled Equal 110 -2.78 0.0064 62.476 -2.87 0.0055 SatterthwaiteUnequal Proc Univariate – Histogram issues V Horecny – Key Insight – Chipsets HTC-M8 (blue) modern chipset HTC-Desire S (Pink) early version chipset Farondeep Bains – Key Insight – Cars and Carparks Richard J Self - University of Derby 11 Amna Al-Mutawa – Key Insight – Time Variability Richard J Self - University of Derby 12 Accuracy? Type of Location Open Rural – most accurate Residential Urban – least accurate Low rise High rise Under car very large errors! Richard J Self - University of Derby 13 Accuracy Variable with Time Consolidated Data – 2420 points Red = > 300m Richard J Self - University of Derby 15 Overall Accuracy of LBS 85% <= 25 metres 2364 out of 2420 (97.6%) <= 500 m Outliers out to 40 to 60 miles! Key Governance Questions What level of accuracy do you need or can you accept? 10m, 50m, 100m, 0.5km, 1km, 10km? What are consequences of uncertain veracity? To your organisation To your customers and clients EU Data Protection regime implications? Consequences of storing when lacking veracity and accuracy? Richard J Self - University of Derby 17 Further Research Replicate the research with a standardised set of parameters and values, based on this year’s exploratory research Control for GPS / Cell based / WiFi / Bluetooth Widen the participation to a world-wide team Extend list of devices / generations / OS / etc. Analyse with IBM’s Watson Analytics (100k data points + needed) – please volunteer!! Extend to High School projects Richard J Self - University of Derby 18