Prediction Overview

Overview

Predictive modeling in Process Mining enables users to analyze historical data and build models that can forecast future outcomes or classify data points. This powerful technique helps organizations make data-driven decisions, optimize processes, and identify potential issues before they occur.

Currently, the System supports two primary types of predictive modeling:

• Classification: Predicts categorical outcomes, such as whether an event will occur or which category a data point belongs to.
• Regression: Predicts continuous numerical values, such as estimating process durations or resource consumption.

Users can select log or CSV files containing data to train models, apply these models to new data and monitor their performance. The predictive models can be deployed for ongoing use, enabling proactive decision-making and process optimization.

Using Predictions

The Predictions menu is located at the end of the Action menu. Users can choose from three options: Train Prediction, Predict, and Prediction Center.

Train Prediction

Training Prediction takes available data and then creates a model from it. To train a Prediction, first select a log or .CSV. The Garage Car Repair data is an easy way to test this feature.

After selecting data and pressing Train Prediction, the following menu appears:

The Activity section contains all the attributes from the log or .CSV file.

Prediction Types

Algorithm

Predict

The Predict action allows the user to apply a prediction model against an imported log or .CSV file. As long as the file has the same characteristics as the model, it will provide a Probability column indicating how the data would perform.

Prediction Center

After creating a prediction model, it can be found within the Prediction Center.

1. The model's name is set when it is created. It can be renamed at any time by right-clicking on the name.
2. The ID field is used to distinguish models from each other at the System level. When calling a model within Decisions, this ID number is critical.
3. The algorithm is determined at the time of creation. See the section above for more information on the different types.
4. The prediction class is the target variable the modeling process estimates during prediction.
5. Predictive models can have one of six different statuses.
   1. Generating: This status will show up while the predictive model is still being trained.

   2. Ready: This status will show up once the generation of the predictive model is completed. 

   3. Deployed: This status will show up for predictive models that are currently ready and deployed.

   4. Ready (Unstable): This status will show up once the predictive model generation is completed, but the model is unstable. A model may be unstable when it is based on a classification algorithm and cannot predict all possible target labels. Quality measures for unstable models are not available.

   5. Deployed (Unstable): This status will show up for unstable predictive models currently ready and deployed.

   6. Failed: This status will show up when the generation of the predictive model fails. 

6. Actions

Icon	Name	Description
	Predictive Model Properties	This opens the Confusion Matrix screen, giving users insight into how the model will function.
	Ready	Changes the status of the model from Ready to Deployed.
	Delete	Deletes the highlighted model. There is no confirmation screen, and this action is permanent. Be careful when using this action.

Confusion Matrix

A Confusion Matrix is a key evaluation tool in predictive modeling, especially for classification tasks. It provides a detailed breakdown of the model’s performance by comparing actual versus predicted outcomes. This matrix helps users understand not just overall accuracy, but also where the model is making errors.

During model training, the System splits the data into 70% for training and 30% for testing. The Confusion Matrix is calculated using this 30% testing dataset. The total number of instances in the matrix equals the sum of False Negatives (FN), False Positives (FP), True Negatives (TN), and True Positives (TP). Values may appear as decimals for multiclass classification, as they are averaged across all classes.

Confusion Matrix for Binary Classification

In binary classification, the Confusion Matrix is a 2x2 table showing how well the model distinguishes between two classes (e.g., "Yes" vs "No", "Positive" vs "Negative"). The matrix is structured as follows:

• True Positive (TP): Model correctly predicts the positive class.
• True Negative (TN): Model correctly predicts the negative class.
• False Positive (FP): The model incorrectly predicts a positive when it is actually negative (Type I error).
• False Negative (FN): The model incorrectly predicts negative when it is actually positive (Type II error).

The sum of all four values equals the total number of test instances. These numbers help compute accuracy, precision, recall, and F1-score metrics.

Confusion Matrix for Multiclass Classification

For multiclass classification, where there are more than two possible outcomes, the Confusion Matrix expands into an n x n table (where n is the number of classes). Each row represents the actual class, and each column represents the predicted class.

In multiclass scenarios, the values in the matrix may be decimal numbers. This is because the system computes averages across all classes to provide a balanced evaluation, especially when class distributions are uneven.

• The diagonal elements (e.g., TP_A, TP_B, TP_C) show correct predictions for each class.
• Off-diagonal elements represent misclassifications between classes.

By analyzing the Confusion Matrix, users can see which classes are most often confused with each other and identify areas for model improvement. Using averaged (decimal) values ensures fair evaluation even in complex, multiclass situations.

How the Confusion Matrix is Displayed in Decisions - Process Mining

The Confusion Matrix is always presented as a 2x2 table, regardless of working with binary or multiclass classification problems. The matrix displays the actual counts of True Positives, True Negatives, False Positives, and False Negatives for binary classification.

Instead of showing a larger matrix for multiclass classification, it calculates and displays the average values for TP, TN, FP, and FN across all classes in the same 2x2 format. This means seeing decimal numbers in the matrix, which represent the mean performance of the model for each class, averaged over all possible classes. This approach provides a clear and consistent summary, making it easy to compare model performance even as the number of classes increases.

This averaged 2x2 confusion matrix helps users quickly understand overall model accuracy and error distribution, without needing to interpret a more complex, multi-dimensional matrix.