K-Means Clustering with K-Centroids

The K-Means Algorithm: A Quick Conceptual Overview

Before we dive into the Alteryx tool, let's briefly revisit how the K-Means algorithm works. Understanding this process will make the tool's configuration options much more intuitive.

K-Means is an iterative algorithm that aims to partition your data into a pre-specified number (K) of distinct, non-overlapping clusters.

To understand its iterative nature, let's read a short explanation.

In short, the process is:

1. Initialization: The algorithm starts by randomly placing K initial points, called centroids, in your data space.
2. Assignment: Each data point is assigned to its closest centroid, typically based on standard Euclidean distance. This creates K initial clusters.
3. Update: The centroid of each cluster is recalculated by taking the mean of all data points assigned to that cluster.
4. Iteration: Steps 2 and 3 are repeated. Data points might get reassigned to different clusters, and the centroids will shift their position. This continues until the cluster assignments no longer change between iterations, a state known as convergence.

Now, let's see how to implement this using Alteryx.

Building the Model with the K-Centroids Cluster Analysis Tool

This tool is the engine of our clustering analysis. A typical clustering workflow in Alteryx involves preparing the data, running the analysis, and then using the results, as shown below. Today, we're focusing on the "Clustering" step.

Let's walk through the configuration of the K-Centroids Cluster Analysis tool.

Configuring the Analysis

To understand each setting, we'll use a combination of short readings and explanations.

As the article mentions, the configuration is straightforward. Let's break it down parameter by parameter.

Based on those readings, here is a summary of the key configuration settings:

1. Solution Name: A unique name for your model object. This is useful for identifying the model later, especially if you build several.

2. Fields: Here, you select the numeric variables you want the algorithm to use for clustering. This is where the work from our previous lesson pays off—you should be selecting only relevant, numeric fields.

3. Standardize the fields: As we covered previously, this is crucial. For most clustering scenarios, you will want to have this checked and use the default z-score method. This ensures all your selected variables are on a level playing field.

4. Clustering method: The tool offers three algorithms. For this lesson and most common use cases, you will select K-Means.

   • K-Means: Uses the mean to define the center of a cluster (the centroid). It's the most widely used clustering algorithm.
   • K-Medians: A variation that uses the median instead of the mean. This makes it less sensitive to outliers.

5. Number of clusters: This is the 'K' in K-Means. You must tell the algorithm how many groups to find. How do you know the right number? That is a key question that we will address in the next lesson. For now, we assume you have a target number in mind (e.g., "I want to find 4 customer segments").

6. Number of starting seeds: Because the algorithm starts by placing centroids randomly, running it once might not give you the best possible result. This setting runs the entire algorithm multiple times (e.g., 10 times), each with a different random starting position. It then automatically keeps the best solution it found. Using at least 10 seeds is good practice to ensure a stable and reliable result.

Understanding the Model Outputs

After you configure and run the tool, it produces two outputs, marked with 'O' and 'R'.

Let's summarize the two outputs:

1. The 'O' (Output) Anchor: The Model Object

• What it is: The actual, serialized model. It's a package containing all the mathematical information about your clusters, including the final centroid locations.
• What it's for: This output is not meant for direct human reading. Its primary purpose is to be fed into another tool, most commonly the Append Cluster tool. The Append Cluster tool takes this model object and your original dataset and adds a new column to your data indicating which cluster each row belongs to.

2. The 'R' (Report) Anchor: The Summary Report

• What it is: A human-readable report that summarizes the clustering results. When you connect a Browse tool to this anchor, you will see tables and charts.
• What it's for: This is where you go to understand and evaluate your clusters. It contains key information like:
  • Cluster Information: A table showing the size of each cluster, the average distance of points from their centroid, and other metrics we will use for evaluation in the next lesson.
  • Cluster Centroids: A table showing the final centroid values for each cluster and each variable. This is extremely useful for creating "personas" or profiles for your clusters (e.g., "Cluster 1 is characterized by high income but low transaction frequency").
  • Biplots: A visual representation of the clusters, which helps in understanding their separation.

Conclusion

In this lesson, you have learned how to take your prepared data and build a K-Means clustering model in Alteryx. We've moved from the theory of data prep to the practical application of the core modeling tool.

Key Takeaways:

• The K-Centroids Cluster Analysis tool is the primary tool for building clustering models in Alteryx.
• The K-Means algorithm is an iterative process that assigns data points to the nearest cluster centroid and then updates the centroids.
• Key configuration steps include selecting the relevant numeric fields, standardizing them with z-score, choosing the K-Means method, and specifying the Number of clusters and Number of starting seeds.
• The tool has two outputs: the 'O' (Object) anchor, which contains the model for use with the Append Cluster tool, and the 'R' (Report) anchor, which provides a human-readable summary for evaluation.

Preview of the Next Lesson:
Today, we had to assume we knew the correct number of clusters to use. But how do you determine if 4 clusters is better than 5? In the next lesson, "Evaluate cluster quality and determine an appropriate number of clusters," we will tackle this question. You will learn how to use the K-Centroids Diagnostics tool and interpret the model report to find the optimal number of clusters and assess how good your final model is.