A new study has revealed a measurement system that detects subway travellers\u2019 movements to help public transport planners make smarter decisions around future investment and management.\nThe study was released as part of the 21st ACM SIGKDD Conference on Knowledge Discovery and Data Mining in Sydney this week, and is titled Traffic Measurement and Route Recommendation System for Mass Rapid Transit (MRT).\nThe system has been operating for several months on Singapore\u2019s Mass Rapid Transit (MRT) network, which has 92 stations and five train lines. It looks at the number of people built up on station platforms, the flows of people making transfers at stations, and the expected travel time for each possible route.\nThe authors of the study, from DataSpark (part of SingTel), developed an algorithm that detects individual subways trips from anonymised real-time location-based data from SingTel.\nAs Singapore\u2019s largest telecommunications company, it was able to tap into the mobile phone location data of 3.98 million customers, which was anonymised ID, latitude, longitude, timestamp and service type.\nPrivacy was taken into account when conducting the study by anonymising the original unique ID for each mobile customer through a two-step non-reversible AES Encryption and Hashing process.\n\u201c[This] makes it impossible to trace back to the original unique ID or the mobile customer,\u201d the authors said.\nThey also complied with Singapore\u2019s Personal Data Protection Act (PDPA), and consent from customers was also given through SingTel\u2019s Data Protection Policy.\nTo ensure that the selected customer base is representative of Singapore\u2019s population, the authors compared the distribution of residences of SingTel customers with Singapore\u2019s 2010 census data. The two distributions highly linearly correlated, meaning it was quite a representative population.\nLooking at the most densely populated stations, which are indoors, the algorithm was able to distinguish SingTel cell towers for the indoor subway from all others when detecting individual trips and their locations.\n\u201cWe define an MRT trip as the sequence of indoor stations s = [s1, . . . sn] a passenger ? passes during their ride and the corresponding timestamps t = [t1, . . . tn] when each of these stations is reached. s1 is the boarding station, sn the station of disembarkation and s2, . . . sn \u2013 1 stations ? passes during their ride.\u201d\nTo test the measurements that the algorithm produced against some \u2018ground truth\u2019 or real measurements, the authors compared with counts taken at popular shopping spot, Orchard MRT station, engaging a market research company to do this.\n\u201cWe determined the number of SingTel subscribers departing and arriving at Orchard MRT station during the same three days on an hourly basis using our MRT trip detection algorithm and market extrapolated these numbers with the scaling factor A. This takes into account both SingTel\u2019s market share in Singapore as well as the mobile device penetration rate in Singapore.\nThe comparison found that there was a strong correlation between the counts and the measurements the algorithm was able to produce.\nOne of the key findings from the study was that travellers do not necessarily always decide to go the shortest route when planning their travel journey.\n\u201cSometimes, the shortest distance may also not be the quickest route due to congestion. As such, the most time efficient route between two MRT stations may be difficult to determine for commuters.\n\u201cWith the implementation of the trip database, we are able to determine the popularity of routes between stations (i.e. the number of people taking any route between two given stations), as well as the average travel time per route.\u201d\nThe current methods of measuring 'crowdedness' and how people flow through a subway network have their shortfalls, the authors said. Video or CCTV and Wi-Fi cannot measure travel times between stations and smart ticketing cards, which give tap on and tap off data points for each traveller, do not capture line transfers.\n\u201cCell phone location data are capable of addressing the information gaps in the approaches above.\u201d\nThe authors are looking to extend their measurement system to also include outdoor trips.