Failure Detection using Driver Behaviour from Telematics

Telematics data have a potential to unlock revenue of 1.5 trillion. Unfortunately this data has not been tapped by many users.

In this case study Karthik Thirumalai would discuss how we can use telematics data to identify driver behaviour and do preventive maintenance in automobile.


Outline/Structure of the Case Study

- Outline of the problem

- Data Analytics Framework

- Exploratory Data Analysis

- Modelling

- Outcomes

Learning Outcome

How to use telematics data to infer driver behaviour and prevent failure using it. The presentation would cover in-depth analysis and visualization of the data. Also explain about different models that were applied to the model and the outcomes of the solution. The case study would touch upon data visualization, exploratory data analysis, machine learning

Target Audience


schedule Submitted 6 months ago

Public Feedback

comment Suggest improvements to the Speaker
  • Dr. Vikas Agrawal
    By Dr. Vikas Agrawal  ~  5 months ago
    reply Reply

    Dear Karthik: Thanks for the proposal. Loved your video on O'Reilly as well. Could you please add a sample of your slides and further details to the structure of the case study, including how much time is spent on each section, and what algorithms and techniques will be shown?

    Warm Regards


    • Karthik Bharadwaj T
      By Karthik Bharadwaj T  ~  4 months ago
      reply Reply

      Hi Vikas,

      Thanks for your message. Regarding the slides, I would have to anonymize client-specific information in it. But the structure of the presentation would be as follows.

      • Business Problem
      • Data Quality challenges
      • Exploratory Data Analysis
        • Spatio temporal visualization on moving vehicles and complaints
        • Understanding data distributions
      • Features Engineering of events indicator related to driving behaviour
      • Model Development
      • Multi class classification
      • Predicting type of failure from driver behaviour Used RF algorithm to solve the problem
      • Model Evaluation

      Would structure the talk with 35 mins for the talk and 10 mins for questions. In the 35 mins would deep dive for the 5-10 mins for business and data understanding and spend the rest on modelling and results. Let me know if the format works for you.

      Thanks, Karthik

  • Anoop Kulkarni
    By Anoop Kulkarni  ~  5 months ago
    reply Reply

    Thanks for your proposal. You indicate use of telematics data, but havent qualified exactly what constitutes "telematics data" for this presentation. Would it be possible to update what type of data, how big, what parameters you intend to use and use it to compare approaches against one another?


    • Karthik Bharadwaj T
      By Karthik Bharadwaj T  ~  5 months ago
      reply Reply
      Hi Anoop, Thanks for asking the data contains 6 months of driver behavioural events from 30K trucks. The events like over speeding, harsh acceleration, coasting have been recorded in the telematics data. We also have complaints information from service station from vehicles. The talk would be on data integration, feature engineering approaches, spatio temporal visualizations, modeling and model validation metrics (precision, recall, ROC) approaches used in solving the problem. - Karthik