Attribute Based Component Forecasting for High Technology Industry using Machine Learning

The Personal System Business Unit, the flagship unit of HP Inc, is powered by some of the most innovative technologies in the industry and has consistently delivered exceptional results. However, the business has been plagued with recurring shortages and over stocking of slow-moving SKUs & Components owing to poor forecast accuracy. The current forecasting framework uses conventional forecasting methods and basic time series models to arrive at the baseline forecast. This approach works well for certain segments and regions with high predictability and noticeable seasonality but fails for areas with erratic demand and weak seasonal behavior. This created a need for HP to develop a robust forecasting approach/framework to improve the accuracy. In this paper, we propose “Attribute based forecasting framework”, a multi model solution, which uses techniques like Text Analytics, Decision Tree, Random Forest, Support Vector Machine and Artificial Neural Networks in building the prediction models.


Outline/Structure of the Case Study

  1. Problem Statement
  2. Current Forecasting Scenario
  3. Personal Systems Portfolio along with Challenges in Forecasting
  4. Model Framework
  5. Results
  6. Future steps

Learning Outcome

  1. The usage of Machine Learning/ Deep Learning techniques for forecasting of rapidly changing technology industry with short life cycle.
  2. Forecasting techniques of dependent variables using independent variables when there is no future information about the latter.
  3. How different attributes of a product can be used in forecasting with irregular or insufficient shipment history

Target Audience

Data Scientist, Forecasters, Demand Planners, Analytics Professional

Prerequisites for Attendees

The attendees should have basic understanding about

  1. Machine learning/ Deep Learning
  2. Time Series Models
  3. Personal Systems(Notebooks & Workstation) Business in General
  4. Demand Planning & Forecasting
schedule Submitted 5 months ago

Public Feedback

comment Suggest improvements to the Speaker
  • Kuldeep Jiwani
    By Kuldeep Jiwani  ~  4 months ago
    reply Reply

    Hi Aswin / Raj,

    You have submitted an interesting topic with good technical content.

    For the program committee to get a better understanding of your talk, can you add more technical details that you would be covering. Like for example your technical journey on this project, why did the traditional time series forecasting models could not achieve more than 55% accuracy. Then what kind of analysis you did to identify the issues and finally came up with an ensemble model that delivers more than 75% accuracy.