Xuncheng Zhang, Timilehin Fasubaa and Michelle Tsai

Introduction

Most people will at some point in their lives have to pay Auto Insurance. However, the heuristics used to decide how much each individual pays are overly generalized. Why should I pay more because I am male? Does the fact that most teenagers drive badly mean I will drive badly and that you should therefore charge me higher? This project seeks to personalize insurance premium costs by looking more closely at individual driving habits and using that extra data to better price insurance. We will be achieving this by getting audio, location and video data of the drivers.

PHAM

  • Problem:
    1. Auto insurers lack specific information about each individual's driving habits, and as a result, resort to using statistics on entire demographic groups
  • Hypothesis:
    1. If cars had extra sensors, such as the black boxes on airplanes, insurance companies will be able to get up to date information about each individual's driving habits. They will know exactly how risky a driver each individual is, and how risky their driving environments are. Insurers will also know whether or not their client is actually at fault during an accident, instead of having to rely on accident reports from police officers, or conflicting accounts from both drivers.
  • Action:
    1. We should add some sensors into cars, and transmit the information back to insurance companies
      • Basic accelerometer, to detect how quickly someone speeds up, and how fast they are driving normally
      • Voice sensors: detects how often the driver converses with passenger
      • Video censor (facing the driver). This will detect how often the driver is distracted from looking at the road, and how quickly they return to paying attention after distractions.
      • GPS: where does the driver and how risky are those areas

  • Metrics:
    1. Insurance companies can create a formula to calculate the overall driving performance, and by using the data they collect by those sensors they can evaluate each individual drivers by computing their total safety scores.
    2. Then we compare the safety scores with the number of accidents of individual users in a period (can be a year), and so on we can figure out the exact insurance discount with different safety scores
  • Evaluation
    1. see if the predicted number of accidents in a year matches the real number of accidents a driver occurs (if the formula is correct)
    2. Compute the extra discount for safety drivers and the extra fine collected from reckless drivers. (compare the gains from this policy and it loss, it in general we lose money, we can adjust discount rate and fine rate )
    3. Do a survey to see if all the users love this policy or not.


LIMITATIONS
  • Privacy: many individuals highly value their privacy, and would not be willing to install these devices in their cars. We will address this concern by letting clients decide to use the black box feature or to decline it. This way, only those who do not feel that the black box is intrusive would be affected by the data change.
  • Price: some users may not be able to afford paying for pricey new sensors. However, any standard smartphone should already contain the relevant censors, and we can simply provide an insurance company app to capture relevant information.
  • Legality: In certain states, passengers must consent to be recorded in two party recording states. We feel that most users of this black box functionality will be private drivers, so they should know their passengers. As a result, there should not be any antagonistic relationships, that would prevent the black box feature from being used. Even if a passenger strongly objected to being filmed, we could simply turn off that sensor, and utilize other sensors like the accelerometer.

Citations