Predictive Modeling Guide

Dacture’s Predictive Modeling flow enables you to achieve two key things:

  1. Predict current expected results based on your data
  2. Model scenarios to forecast the impact of potential changes

As an example:

  1. You can predict your current expected sign-up rate
  2. Then build a scenario where you predict the sign-up rate if more people start consuming specific pieces of content or interact with certain parts of the site

Model Training

Before making a prediction you have to have a trained model. If you’ve already trained a model, you can skip this section.

To train a model:

  1. Choose your data set. If you haven’t uploaded a data set or connected Dacture to your data source, you’ll need to do that.

  2. Choose what you want to predict (e.g., sign-ups; converting to paid). What you are able to predict depends on the data you bring.

  3. If your data set has an identifier column, you can optionally choose to identify it here.

  4. Remove columns that are
    • Unknown at the time your predicted variable is collected

      For example, the column “reason for churn” would be unknown when you are trying to predict customer churn because you only collect this information after a customer has churned; therefore “reason for churn” should be dropped before you train your model.

    • Redundant

      For example, “time spent on tutorial (hours)” and “time spent on tutorial (minutes)” are redundant with one another; therefore, you can remove one of these predictors.

    • Irrelevant

      For example, you may have a column for “IP address” which may be irrelevant for predicting customer churn; therefore, you can drop “IP address” before training your model.

  5. Click Train Model.

    If your data set is imbalanced, you’ll be prompted to choose if you want us to balance it. Balancing the data set creates more accurate predictions.

    The training time is dependent on the size of the data set. You can continue using Dacture while the model trains

    When the model is done training, Dacture displays a chart showing the most impactful predictors.

Predicting Current Expected Results

To make a prediction for your current expected results:

  1. Train a model or open a previously saved trained model.

  2. Upload a data set for prediction. This can be the same CSV you used for training the model or a new CSV with the same columns, but different data. As an example, you could have trained a model using previous user sessions, and are now running predictions based on the most recent sessions to determine their expected outcome.

  3. Click Predict.

    Your results are displayed with details about the prediction.

Scenario Modeling

Scenario modeling or “what-if” prediction helps you determine the impact of potential changes before you make them. You can use this in partnership with predicting the current expected outcome of something so that you can compare results.

Use the results in the Most Impactful Predictors chart as levers in your scenario models.

As an example, you can run an initial prediction and see that as things stand currently, the sign-up rate would be 5%. You can then model a scenario where you change the value of one or more events and see that the change would lead to an expected sign-up rate of 7%.

To model a scenario:

  1. Click What If
  2. Add one or more events and adjust their values as desired.
  3. Click What If

Make additional predictions and adjustments based on your results.