Predicting safety incidents using machine learning

Patent metadata

Field	Value
Publication number	US-10720050-B2
Application number	US-201615297050-A
Country	US
Kind code	B2
Filing date	Oct 18, 2016
Priority date	Oct 18, 2016
Publication date	Jul 21, 2020
Grant date	Jul 21, 2020

Abstract

A safety system associated with a travel coordination system collects safety data describing safety incidents by providers and generates a plurality of safety incident prediction models using the safety data. The safety incident prediction models predict likelihoods that providers in the computerized travel coordination system will be involved in safety incidents. Two types of safety incidents predicted by the safety system include dangerous driving incidents and interpersonal conflict incidents. The safety system uses the plurality of safety incident prediction models to generate a set of predictions indicating probabilities that a given provider in the travel coordination system will be involved in a safety incident in the future. The safety system selects a safety intervention for the given provider responsive to the set of predictions and performs the selected safety intervention on the given provider.

First claim

What is claimed is:

1. A computer-implemented method comprising:
collecting trip data associated with trips by providers of a computerized travel coordination system, the trip data including trips that have safety incidents, wherein safety incidents include dangerous driving incidents and interpersonal conflicts, and further including trips that do not have safety incidents;
generating a plurality of safety incident prediction models using the trip data, the safety incident prediction models predicting likelihoods that providers of the computerized travel coordination system will be involved in safety incidents, wherein generating the plurality of safety incident prediction models comprises:
obtaining a set of training data from the collected trip data;
adjusting the set of training data by randomly removing data about trips that do not have safety incidents from the set of training data to generate an adjusted training set that includes a specified ratio of trip data for trips that have safety incidents to trip data for trips that do not have safety incidents;
generating, for each of a set of multiple specified timeframes, a dangerous driving incident prediction model for determining probabilities that providers will be involved in dangerous driving incidents within the specified timeframe;
training the generated dangerous driving incident prediction models with the adjusted training set using response variables that are indicative of the occurrence of dangerous driving incidents;
generating, for each of a set of multiple specified timeframes, an interpersonal conflict incident prediction model for determining probabilities that providers will be involved in interpersonal conflict incidents within the specified timeframe; and
training the generated interpersonal conflict incident prediction models with the adjusted training set using response variables that are indicative of the occurrence of interpersonal conflicts;

generating a set of predictions indicating probabilities that a given provider of the computerized travel coordination system will be involved in a safety incident in the future using the plurality of safety incident prediction models comprising the trained dangerous driving incident prediction models and the trained interpersonal conflict incident prediction models;
selecting a safety intervention for the given provider responsive to the set of predictions; and
performing the selected safety intervention on the given provider.

2. The computer-implemented method of claim 1 , wherein collecting trip data comprises:
collecting provider level predictors relating to the providers' quality of controlling vehicles carrying riders; and
collecting city level predictors relating to likelihoods that safety incidents will occur in particular geographical areas;
wherein the plurality of safety incident prediction models are generated using the provider level predictors and the city level predictors.

3. The computer-implemented method of claim 1 , wherein selecting the safety intervention for the given provider responsive to the set of predictions comprises:
identifying a set of potential safety interventions for the provider;
assigning an impact score to each potential safety intervention in the identified set; and
selecting the safety intervention for the given provider responsive to the assigned impact scores and the probabilities that the given provider will be involved in a safety incident in the future.

4. The computer-implemented method of claim 3 , wherein selecting the safety intervention for the given provider responsive to the assigned impact scores comprises:
mapping the set of potential safety interventions to ranges of probabilities, with potential interventions having comparatively higher impact scores mapped to comparatively higher probabilities of a safety incident occurring; and
selecting the safety intervention responsive to the mapping.

5. The computer-implemented method of claim 1 , wherein generating the plurality of safety incident prediction models using the trip data comprises:
training the safety incident prediction models using supervised machine learning.

6. A computer system comprising:
a computer processor for executing computer program instructions; and
a non-transitory computer-readable storage medium storing instructions executable by the processor to perform steps comprising:
collecting trip data associated with trips by providers in a computerized travel coordination system, the trip data including trips that have safety incidents, wherein safety incidents include dangerous driving incidents and interpersonal conflicts, and further including trips that do not have safety incidents;
generating a plurality of safety incident prediction models using the trip data, the safety incident prediction models predicting likelihoods that providers in the computerized travel coordination system will be involved in safety incidents, wherein generating the plurality of safety incident prediction models comprises:
obtaining a set of training data from the collected trip data;
adjusting the set of training data by randomly removing data about trips that do not have safety incidents from the set of training data to generate an adjusted training set that includes a specified ratio of trip data for trips that have safety incidents to trip data for trips that do not have safety incidents;
generating, for each of a set of multiple specified timeframes, a dangerous driving incident prediction model for determining probabilities that providers will be involved in dangerous driving incidents within the specified timeframe;
training the generated dangerous driving incident prediction models with the adjusted training set using response variables that are indicative of the occurrence of dangerous driving incidents;
generating, for each of a set of multiple specified timeframes, an interpersonal conflict incident prediction model for determining probabilities that providers will be involved in interpersonal conflict incidents within the specified timeframe; and
training the generated interpersonal conflict incident prediction models with the adjusted training set using response variables that are indicative of the occurrence of interpersonal conflicts;

generating a set of predictions indicating probabilities that a given provider in the computerized travel coordination system will be involved in a safety incident in the future using the plurality of safety incident prediction models comprising the trained dangerous driving incident prediction models and the trained interpersonal conflict incident prediction models;
selecting a safety intervention for the given provider responsive to the set of predictions; and
performing the selected safety intervention on the given provider.

7. The computer system of claim 6 , wherein collecting trip data comprises:
collecting provider level predictors relating to the providers' quality of controlling vehicles carrying riders; and
collecting city level predictors relating to likelihoods that safety incidents will occur in particular geographical areas;
wherein the plurality of safety incident prediction models are generated using the provider level predictors and the city level predictors.

8. The computer system of claim 6 , wherein selecting the safety intervention for the given provider responsive to the set of predictions comprises:
identifying a set of potential safety interventions for the provider;
assigning an impact score to each potential safety intervention in the identified set; and
selecting the safety intervention for the given provider responsive to the assigned impact scores and the probabilities that the given provider will be involved in a safety

Assignees

• Uber Technologies Inc

Inventors

• Jeon Sangick

Classifications

• G06N20/00 Primary

  Machine learning · CPC title

• G08G1/0141 Primary

  for traffic information dissemination · CPC title

• G08G1/20

  Monitoring the location of vehicles belonging to a group, e.g. fleet of vehicles, countable or determined number of vehicles · CPC title

• G08G1/012

  from other sources than vehicle or roadside beacons, e.g. mobile networks · CPC title

• G08G1/202

  Dispatching vehicles on the basis of a location, e.g. taxi dispatching · CPC title