Identifying high risk trips using continuous call sequence analysis

1. A system for identifying a high risk trip, the system comprising:

one or more processors; and

a memory storing instructions that, when executed by the one or more processors, cause the system to perform:

training a machine learning model based on a plurality of training trips sampled from a plurality of historical trips to determine a risk for a given trip request, wherein the training comprises:

generating a plurality of historical scores for each of the plurality of training trips based on features of the each training trip;

determining a plurality of historical patterns based on the plurality of historical scores for each of the plurality of training trips;

inputting the plurality of historical patterns into the machine learning model to obtain predictions;

adjusting parameters of the machine learning model to converge the predictions and corresponding trip labels of the plurality of training trips, wherein the parameters are adjusted to penalize incorrect predictions and thus minimize future incorrect predictions, and the parameters are adjusted with a less degree for a false-positive prediction than for a false-negative prediction;

receiving a current trip request by a passenger, wherein the passenger is associated with a plurality of prior trip requests within a prior time window, wherein each of the plurality of prior trip requests is associated with a plurality of features comprising a plurality of time window-related features, a plurality of prior driver features, a plurality of passenger features, and a prior trip outcome, and wherein the prior trip outcome indicates a positive trip outcome status or a negative outcome status;

generating a plurality of scores for each of the plurality of prior trip requests from the plurality of features associated with the prior trip request;

determining a plurality of patterns from the corresponding scores of the plurality of prior trip requests using a time series technique;

determining a risk for the current trip request from the plurality of patterns by inputting the plurality of patterns into the machine learning model and obtaining a predicted risk of the current trip from the machine learning model; and

determining the risk for the current trip request is below a trip risk threshold and assigning a driver to the current trip request.

2. The system of claim 1 , wherein the negative trip is an occurrence of an incident associated with a prior trip request, and wherein the positive trip is an absence of an incident associated with a prior trip request.

3. A system for identifying a high risk trip, the system comprising:

one or more processors; and

a memory storing instructions that, when executed by the one or more processors, cause the system to perform:

training a machine learning model based on a plurality of training trips sampled from a plurality of historical trips to determine a risk for a given trip request, wherein the training comprises:

generating a plurality of historical scores for each of the plurality of training trips based on features of the each training trip;

determining a plurality of historical patterns based on the plurality of historical scores for each of the plurality of training trips;

inputting the plurality of historical patterns into the machine learning model to obtain predictions;

adjusting parameters of the machine learning model to converge the predictions and corresponding trip labels of the plurality of training trips, wherein the parameters are adjusted to penalize incorrect predictions and thus minimize future incorrect predictions, and the parameters are adjusted with a less degree for a false-positive prediction than for a false-negative prediction;

receiving a current trip request by a passenger, wherein the passenger is associated with a plurality of prior trip requests within a prior time window, wherein each of the plurality of prior trip requests is associated with a plurality of features comprising a time window profile, a prior driver profile, a passenger profile, and a prior trip outcome;

generating a plurality of scores for each of the plurality of prior trip requests from the plurality of features associated with the prior trip request;

determining a plurality of patterns from the corresponding scores of the plurality of prior trip requests;

determining a risk for the current trip request from the plurality of patterns by inputting the plurality of patterns into the machine learning model and obtaining a predicted risk of the current trip from the machine learning model; and

assigning a current driver to the current trip request based on the risk of the current trip request relative to a trip risk threshold.

4. The system of claim 3 , wherein the plurality of features includes a number of drivers assigned to the plurality of prior trip requests, a number of the plurality of prior trip requests, trip destinations of the plurality of prior trip requests, trip request time of the plurality of prior trip requests, and a prior trip request of the plurality of prior trip requests having a trip completion time after a trip request time of a subsequent prior trip request of the prior trip requests.

5. The system of claim 3 , wherein the prior driver profile of the prior trip request includes an age of a prior driver assigned to the prior trip request, a gender of the prior driver, a rating of the prior driver, a presence of a photo of the driver, and a car model associated with a profile of the driver.

6. The system of claim 3 , wherein the passenger profile includes a behavior history of the passenger and stored payment information of the passenger.

7. The system of claim 3 , wherein determining the plurality of patterns includes determining the plurality of patterns from the corresponding scores of the plurality of prior trip requests using a time series technique.

8. The system of claim 3 , wherein the plurality of training trips are sampled from an original dataset, and wherein the plurality of training trips have a higher occurrence rate of training trip requests with positive trip than the original dataset.

9. The system of claim 3 , wherein assigning the current driver to the current trip request comprises determining the risk of the current trip request is below the trip risk threshold.

10. A method for identifying a high risk trip, the method comprising:

under control of one or more processors:

training a machine learning model based on a plurality of training trips sampled from a plurality of historical trips to determine a risk for a given trip request, wherein the training comprises:

generating a plurality of historical scores for each of the plurality of training trips based on features of the each training trip;

determining a plurality of historical patterns based on the plurality of historical scores for each of the plurality of training trips;

inputting the plurality of historical patterns into the machine learning model to obtain predictions;

adjusting parameters of the machine learning model to converge the predictions and corresponding trip labels of the plurality of training trips, wherein the parameters are adjusted to penalize incorrect predictions and thus minimize future incorrect predictions, and the parameters are adjusted with a less degree for a false-positive prediction than for a false-negative prediction;

receiving a current trip request by a passenger, wherein the passenger is associated with a plurality of prior trip requests within a prior time window, wherein each of the plurality of prior trip requests is associated with a plurality of features comprising a time window profile, a prior driver profile, a passenger profile, and a prior trip outcome;

generating a plurality of scores for each of the plurality of prior trip requests from the plurality of features associated with the prior trip request;

determining a plurality of patterns from the corresponding scores of the plurality of prior trip requests;

determining a risk for the current trip request from the plurality of patterns by inputting the plurality of patterns into the machine learning model and obtaining a predicted risk of the current trip from the machine learning model; and

assigning a current driver to the current trip request based on the risk of the current trip request relative to a trip risk threshold.

11. The method of claim 10 , wherein the plurality of features includes a number of drivers assigned to the plurality of prior trip requests, a number of the plurality of prior trip requests, trip destinations of the plurality of prior trip requests, trip request time of the plurality of prior trip requests, and a prior trip request of the plurality of prior trip requests having a trip completion time after a trip request time of a subsequent prior trip request of the prior trip requests.

12. The method of claim 10 , wherein the prior driver profile of the prior trip request includes an age of a prior driver assigned to the prior trip request, a gender of the prior driver, a rating of the prior driver, a presence of a photo of the driver, and a car model associated with a profile of the driver.

13. The method of claim 10 , wherein the passenger profile includes a behavior history of the passenger and stored payment information of the passenger.

14. The method of claim 10 , wherein determining the plurality of patterns includes determining the plurality of patterns from the corresponding scores of the plurality of prior trip requests using a time series technique.

15. The method of claim 10 , wherein the plurality of training trips are sampled from an original dataset, and wherein the plurality of training trips have a higher occurrence rate of training trip requests with positive trip than the original dataset.

16. The method of claim 10 , wherein assigning the current driver to the current trip request comprises determining the risk of the current trip request is below the trip risk threshold.