Creating a One AI Voluntary Attrition Risk Model with Video

This is a step-by-step guide for creating a predictive voluntary attrition risk model in One AI. A predictive voluntary attrition risk model is the most common machine learning model in One AI.  Please ensure you have an understanding of machine learning in One AI by reviewing our Introduction to Machine Learning in One AI  guide and/or our One AI Recipes Overview video.

Jump to the video - Creating a Voluntary Attrition Risk ML Model in One AI

Step 1. Configuring your augmentation

One AI automatically selects default settings for most things, so most of the configuration for machine learning relates to data framing.  Data framing has been a guided process since the release of One AI Recipes.  

In the following steps, you will be creating a new augmentation and configuring it using the Voluntary Attrition Risk One AI Recipe.

  1. From the Data menu in the navigation bar select Augmentations.
  2. Click the + Add Augmentation button on the right.

  1. Give the Augmentation a name. Please note that Augmentations cannot be re-named once saved.

  1. From the Augmentation Type dropdown, select Machine Learning Augmentations > Classification.
    Note: Although recipes for different types of Augmentations will be released in the future, all existing recipes are for Classification problems.

  1. Select the Augmentation Query option and click Configure Augmentation Query.

  1. This will bring you to the Configure Augmentation Query page. 

  1. In the What are you interested in predicting? dropdown menu, choose Voluntary Attrition. The description tells you what One AI is going to help you predict.
    Note: Until you make selections in a recipe, you can switch between recipes. Once you begin making selections, you can only switch to Custom / Advanced Model.

  1. Scroll down the page and answer each of the questions using the inline help for guidance. Note that some questions will appear as you complete the series of questions;                                                
    1. Which column or metric would you like to use to define Voluntary Terminations?
      • Column or Metric? : Metric
      • Select a Metric : Terminations - Voluntary
      • Filters : None
    2. What Headcount do you wish to make predictions for?
      • Metric : EOP Headcount
      • Unique Identifier : one.employee.person_id
      • Population Date : Today
      • Filters : None
    3. From your Headcount date, how far into the future do you want to predict?
      • Time Interval : 1 Year(s)
    4. How much history do you want to use to train your predictive model?
      • Number of Training Intervals : 1
    5. Since you're using a metric to define Voluntary Terminations, would you like to override the values? 
      • Here you will override the values rather than accepting the default. 
        • Click on Load Target Metric Values
        • Null Value : No Termination
        • All other values : Termination
      • Positive Label : Termination
    6. Which core attributes do you want to use in your prediction?
      • By default, a narrow scope is applied.  This means all columns in the table the headcount metric is created from are included.  Leave this selection applied.
    7. Which generative attributes do you want to use in your prediction?
      • None
    8. Would you like to verify that all of the selections you have made are valid?
      • Click on Generate Data Statistics
      • Wait for it to process and be sure to see Success at the top before saving.
  2. Click the Save icon at the top right of your screen to save your selections and bring you back to the Augmentation Configuration menu.
  3. Scroll to the bottom of Augmentation Configuration and then click Create.
    Do not forget this step or you will lose all of the selections you made.
  4. Congratulations! Your augmentation recipe has been created.  The next step is to run your augmentation. 

Step 2. Run your augmentation

Running an augmentation executes a series of processes that structure and clean data then select the best predictive model by testing multiple algorithm and parameter combinations.  The resulting model is then used to make predictions.  

Running an augmentation is a simple act for you to perform but results in a lot of work being performed by One AI.  For this reason, runs can take over an hour to complete.

  1. Find the augmentation you just created on the Augmentations page. 
  2. Click on the Run button for your augmentation.  

  1. A warning asking if you are sure you want to start a new run will appear;. 

  1. Select Confirm to proceed or Cancel to go back.
  2. The augmentation will now begin running.  
  3. It can take over an hour for the run to complete as One AI’s machine learning pipeline builds many models then chooses the best. Check the status of the run by selecting Runs for your augmentation and noting the status listed.   
  4. The Status progression is as follows:
    Preparing Run Data > Teaching Machines > Pending > Deployed (more about this status later)
  5. There is also an option to manually Cancel the run. 
  6. Once the augmentation reaches the Pending status, a model has been created and run against the prediction data set.  For an explanation of what this means please see the Introduction to Machine Learning in One AI article.  
  7. Once your run has reached Pending, then you can view your augmentation results. 

Step 3. Review your augmentation results

One AI provides a number of data and visualizations that allow you to see how your predictive model performed.  It often takes multiple cycles of configuration, machine learning, and validation to get to a satisfactory result.  

Once the augmentation reaches the Pending status, a model has been created and run against the prediction data set.  For an explanation of what this means please see the Introduction to Machine Learning in One AI article.  At this point the results are available to be reviewed.

  1. Find your augmentation on the Augmentations page 
  2. Click on Runs
  3. Select the run in Pending status 
  4. You’re now viewing the Exploratory Data Analysis (EDA) Report; 
    • In the Overview section, the Number of variables (Attributes), Number of observations and the Variable types are worth noting.
    • The Variable Status section is very important because this is where the attributes that were selected as features are listed along with the attributes that were not selected and  the reasons why selections were or were not made.
    • Clicking on any feature will bring you to an analysis of that feature.
    • Selecting the Toggle details option for that feature expands a number of analytic views of the data.  Worth noting is that each view can be filtered to either of the labels (outcomes) by using the v icon in the headers.
    • For more information about the EDA Report, please read the EDA Report Introduction.
  5. Select the Results Summary tab at the top or View Results Summary button in the footer.  The Results Summary report provides a large amount of detail about the model and how it performed.
    • Review the Estimation Details.  
      • Which algorithm was selected?
      • Was upsampling employed?  
      • Was the positive label correctly defined?
      • How many predictions are there for each of the labels?
    • Scrolling down, note the Feature Importances.  Which are the most important and by what magnitude?
    • Below the Feature Analysis section is the Classification Report.  This section contains metrics about how well the model performed.  This information is essential as a poorly performing model does not generate trustworthy predictions. Precision, recall, and the f1 score are the most widely used performance measures in One AI.  To learn more about these measures see the One AI FAQs article. 
    • For more information about the Results Summary Report, please see One AI Modeling Report Introduction.

The process of creating a predictive model can be completed in under a day, including run and review time.  While custom models are possible, One AI has standardized much of the configuration for common scenarios like voluntary attrition risk and created standardized Recipes designed to assist you in creating, running, and reviewing augmentations.

Once you have created a voluntary attrition risk model, the next step is to refine it. Learn more about strategies for refinement in the Refining a Machine Learning Model in One AI article.

Video - Creating a One AI Voluntary Attrition Risk Model

Watch this in-depth video to learn how to create and deploy predictive Voluntary Attrition Risk models, also known as flight risk models, for your organization.

You will learn:

  • What a voluntary attrition risk model does
  • How to configuring a new machine learning model in One AI
    • How to select a headcount population for your model
    • How to set a prediction timeframe and historical data timeframe
    • How to select core attributes for the model
    • How to select generative attributes for the model
  • How to validate your selections in the One AI query builder and download reports of the model's data

Run time: >17 minutes 

Was this article helpful?

0 out of 0 found this helpful

Comments

0 comments

Please sign in to leave a comment.