Analyzing Individual Driver’s Performance
Analyzing Individual Driver’s Performance
Based on the separated digital traces for each individualdriver, we can further extract the passenger-delivery traces ofeach driver according to the passenger status (i.e., “occupied”or “vacant”). We can then estimate the taxi revenue in eachtime slot by considering the accumulated passenger-deliverydistance. By averaging the revenue in each time slot during along time period (e.g., a month), we are able to measure the per-formance of a driver. In this section, we will investigate the per-formance of all the taxi drivers in different time slots and givean empirical study on the factors influencing the performance.Taxi Performance Quantification:First, we split a dayinto five time slots:late night(00:00–05:59),morning(06:00–09:59),noon(10:00–13:59),afternoon(14:00–17:59), andevening(18:00–23:59). Since workdays have quite differenttaxi service patterns from weekends and holidays, in this paper,we only use the taxi GPS traces in workdays to illustrate ourideas. The performance of a taxi is measured by hourly revenuewithin each time slot. Since taxi drivers may only work part ofthe time within a time slot, for example, one night shift drivermight work until 4:00, another driver might work until 06:20and hand over the taxi to the partner (exceeding 20 min in themorning time slot), we only count the taxis that serve more thanhalf of the time slot and discard those taxi GPS traces that serveless than half of the time slot.The distribution of taxi drivers’ performance in differenttime slots is shown in Fig. 4. We can observe that, gener-ally speaking, taxi drivers perform best in thenoontime slot(around 30–45 RMB/h) and worst inlate nighttime slot (around5–20 RMB/h); they also have better performance in theeveningandafternoontime slots than themorningtime slot. Thedistribution of taxi driver performance in each time slot roughlyfollows a normal distribution, which is compliant with theobservation .Call Taxi in Kumbakonam
Since we aim to investigate taxi service strategies in termsof drivers’ performance, we first investigate the consistency ofindividual driver’s performance over different time slots. Peoplemight think that good taxi drivers perform well in all the timeslots and vice versa. Through our study, it is found that only110 taxis performing best in the late night time slot are amongthe top 500 taxis performing well during the nighttime (con-taining both theeveningandlate nighttime slots), and only 67taxis performing best in the daytime are among the top 500 taxisif we count the performance in the three time slots containingmorning,noon, andevening. Our study shows that less than22% of the taxi drivers consistently perform well in the night-time and only 13.4% in the daytime. Thus, unlike the methodproposed in [14], which chooses good/ordinary taxis accordingto the daily revenue, we investigate taxi drivers’ performancein each time slot and expect to obtain more reasonable andaccurate results.For each time slot, we select the 500 top-performing taxisas good samples and 500 taxis in the middle range (around themean of the normal distributions in Fig. 4) as ordinary taxis.The reason why we do not select the taxis in the bottom rangeis that many factors might cause taxis’ extra low performanceand these factors are irrelevant to taxi service strategies. Bycomparing good taxis with ordinary ones, we can discover somesimple taxi service patterns that are closely related to revenue.In the following section, we will report a brief empirical studyon two influencing factors.2) Number of Passenger Delivery Trips:Intuitively, a largernumber of passenger-delivery trips should give a higher overallrevenue. We use the correlation coefficient between the numberof passenger-delivery tripsPand revenueRto validate this intuition, wherepiandriare the total numberof passenger-delivery trips and the revenue of the taxi driveri, respectively, andndenotes the number of taxi drivers. Theresults for all the time slots are shown in Table I. The correlationcoefficient in most time slots is close to 1, suggesting thatthe revenue is highly correlated to the number of passenger-delivery trips.We also compare the average number of passenger-deliverytrips for good and ordinary taxis in Table II in all time slots.The number of passenger-delivery trips per hour is displayed as“Mean±Std.” We can see that good taxi drivers complete moretrips than ordinary ones, where the difference is at least 21% inthe afternoon time slot and at most 87% in the nighttime slot.These results imply that good drivers are always more efficientin finding next passengers than ordinary ones.3) Popular Passenger Pickup and Drop-Off Areas:Anotherstraightforward but important influencing factor on taxi drivers’performance is the taxi operation area. This factor has beenconsidered in a number of studies [4], [13], [21] to guidetaxi drivers to find passengers. In Fig. 5, we plot the top 99pickup and drop-off hotspots at different time slots. We cansee that the railway station and Zhejiang University are amongthe popular pickup and drop-off areas across all the time slots.Residential areas become popular at night as many people arereturning home. The pickup number decreases greatly from thedowntown to suburb areas. It is noted that the numbers in thegrid cells in Fig. 5 refer to the ranking of the area in termsof the number of pickup or drop-off events. Apparently, thesame grid cell may have different numbers in different timeslots (i.e., night, morning, noon, afternoon, and evening) dueto the variation of the taxi and passenger demand
https://srisivasakthitravels.com/
https://goo.gl/maps/1rNVoifRFQ9snSor6
Based on the separated digital traces for each individualdriver, we can further extract the passenger-delivery traces ofeach driver according to the passenger status (i.e., “occupied”or “vacant”). We can then estimate the taxi revenue in eachtime slot by considering the accumulated passenger-deliverydistance. By averaging the revenue in each time slot during along time period (e.g., a month), we are able to measure the per-formance of a driver. In this section, we will investigate the per-formance of all the taxi drivers in different time slots and givean empirical study on the factors influencing the performance.Taxi Performance Quantification:First, we split a dayinto five time slots:late night(00:00–05:59),morning(06:00–09:59),noon(10:00–13:59),afternoon(14:00–17:59), andevening(18:00–23:59). Since workdays have quite differenttaxi service patterns from weekends and holidays, in this paper,we only use the taxi GPS traces in workdays to illustrate ourideas. The performance of a taxi is measured by hourly revenuewithin each time slot. Since taxi drivers may only work part ofthe time within a time slot, for example, one night shift drivermight work until 4:00, another driver might work until 06:20and hand over the taxi to the partner (exceeding 20 min in themorning time slot), we only count the taxis that serve more thanhalf of the time slot and discard those taxi GPS traces that serveless than half of the time slot.The distribution of taxi drivers’ performance in differenttime slots is shown in Fig. 4. We can observe that, gener-ally speaking, taxi drivers perform best in thenoontime slot(around 30–45 RMB/h) and worst inlate nighttime slot (around5–20 RMB/h); they also have better performance in theeveningandafternoontime slots than themorningtime slot. Thedistribution of taxi driver performance in each time slot roughlyfollows a normal distribution, which is compliant with theobservation .Call Taxi in Kumbakonam
Since we aim to investigate taxi service strategies in termsof drivers’ performance, we first investigate the consistency ofindividual driver’s performance over different time slots. Peoplemight think that good taxi drivers perform well in all the timeslots and vice versa. Through our study, it is found that only110 taxis performing best in the late night time slot are amongthe top 500 taxis performing well during the nighttime (con-taining both theeveningandlate nighttime slots), and only 67taxis performing best in the daytime are among the top 500 taxisif we count the performance in the three time slots containingmorning,noon, andevening. Our study shows that less than22% of the taxi drivers consistently perform well in the night-time and only 13.4% in the daytime. Thus, unlike the methodproposed in [14], which chooses good/ordinary taxis accordingto the daily revenue, we investigate taxi drivers’ performancein each time slot and expect to obtain more reasonable andaccurate results.For each time slot, we select the 500 top-performing taxisas good samples and 500 taxis in the middle range (around themean of the normal distributions in Fig. 4) as ordinary taxis.The reason why we do not select the taxis in the bottom rangeis that many factors might cause taxis’ extra low performanceand these factors are irrelevant to taxi service strategies. Bycomparing good taxis with ordinary ones, we can discover somesimple taxi service patterns that are closely related to revenue.In the following section, we will report a brief empirical studyon two influencing factors.2) Number of Passenger Delivery Trips:Intuitively, a largernumber of passenger-delivery trips should give a higher overallrevenue. We use the correlation coefficient between the numberof passenger-delivery tripsPand revenueRto validate this intuition, wherepiandriare the total numberof passenger-delivery trips and the revenue of the taxi driveri, respectively, andndenotes the number of taxi drivers. Theresults for all the time slots are shown in Table I. The correlationcoefficient in most time slots is close to 1, suggesting thatthe revenue is highly correlated to the number of passenger-delivery trips.We also compare the average number of passenger-deliverytrips for good and ordinary taxis in Table II in all time slots.The number of passenger-delivery trips per hour is displayed as“Mean±Std.” We can see that good taxi drivers complete moretrips than ordinary ones, where the difference is at least 21% inthe afternoon time slot and at most 87% in the nighttime slot.These results imply that good drivers are always more efficientin finding next passengers than ordinary ones.3) Popular Passenger Pickup and Drop-Off Areas:Anotherstraightforward but important influencing factor on taxi drivers’performance is the taxi operation area. This factor has beenconsidered in a number of studies [4], [13], [21] to guidetaxi drivers to find passengers. In Fig. 5, we plot the top 99pickup and drop-off hotspots at different time slots. We cansee that the railway station and Zhejiang University are amongthe popular pickup and drop-off areas across all the time slots.Residential areas become popular at night as many people arereturning home. The pickup number decreases greatly from thedowntown to suburb areas. It is noted that the numbers in thegrid cells in Fig. 5 refer to the ranking of the area in termsof the number of pickup or drop-off events. Apparently, thesame grid cell may have different numbers in different timeslots (i.e., night, morning, noon, afternoon, and evening) dueto the variation of the taxi and passenger demand
https://srisivasakthitravels.com/
https://goo.gl/maps/1rNVoifRFQ9snSor6
Comments
Post a Comment