Data Preparation for Clustering

Part 1: Selecting the Right Fields for Clustering

The first step in preparing your data is deciding which information, or which fields, to include in your analysis. A poor choice of fields will lead to meaningless clusters. There are two main considerations here: the data type and the business relevance.

1. Use Only Numeric Fields

Clustering algorithms like K-Means (which we'll use soon) work by measuring the mathematical distance between data points. To do this, they require numeric data. You cannot calculate the "distance" between non-numeric categories like 'Red' and 'Blue' or between text fields like customer names.

Therefore, you must select fields with numeric data types, such as integers or floating-point numbers (doubles, floats).

If you have a relevant categorical field (e.g., a customer tier like 'Gold', 'Silver', 'Bronze'), you would need to convert it into a numeric format first, for example through one-hot encoding. However, for the scope of the Alteryx Advanced certification, your focus will be primarily on using existing numeric fields.

2. Ensure Business Relevance

Simply choosing all available numeric fields is not a good strategy. The fields you select should be directly related to the business question you are trying to answer.

For example, if your goal is to segment customers based on their purchasing behavior, you should select fields like:

• TotalSales
• AverageOrderValue
• NumberOfTransactions
• DaysSinceLastPurchase

Including irrelevant fields, such as CustomerID or a numeric PostalCode, would confuse the model. A CustomerID is just an identifier and has no mathematical meaning, while a PostalCode doesn't represent a continuous quantity and would skew the distance calculations unless treated very carefully.

Think about it: If you wanted to group different types of cars, would you use TopSpeed and EngineSize, or LicensePlateNumber and Color? The first two are relevant numeric features for performance-based grouping; the latter two are not.

Part 2: The Importance of Feature Scaling

Once you've selected your relevant numeric fields, you'll likely run into a common problem: your variables have different scales.

Imagine you're clustering customers using two fields:

• Age: ranging from 20 to 75
• Annual Income: ranging from $30,000 to $250,000

The range of Annual Income is vastly larger than the range of Age. When a clustering algorithm calculates the distance between two customers, the difference in income will completely dominate the calculation. The difference in age will become almost irrelevant. This will prevent the algorithm from finding meaningful patterns that involve both variables.

To solve this, we use feature scaling—a process that transforms our data to put all variables on a comparable scale. Let's look at a brief article that introduces this concept.

Part 3: How to Scale Your Data

There are two primary methods for scaling data that you need to know for clustering in Alteryx. Both are conveniently built directly into the K-Centroids Cluster Analysis tool.

Let's read about these two methods before we see how they appear in Alteryx.

As the article explained, here are the two key methods:

1. Standardization (Z-score)

• What it does: Transforms the data so that it has a mean of 0 and a standard deviation of 1.
• How it works: It subtracts the mean from each data point and then divides by the standard deviation.
• When to use it: This is the most common and generally recommended method for clustering. It is less sensitive to outliers than normalization and works well with algorithms that assume your data is normally distributed (even though K-Means does not strictly require this).

2. Normalization (Unit Interval / Min-Max Scaling)

• What it does: Rescales the data to a fixed range, usually [0, 1]. The smallest value becomes 0, the largest becomes 1, and all other values fall somewhere in between.
• How it works: It subtracts the minimum value from each data point and then divides by the range (max value - min value).
• When to use it: This method is useful but can be heavily influenced by outliers. If you have one extremely high or low value, it can squash all the other data points into a very small range, which might hide their underlying patterns.

Applying Scaling in Alteryx

Fortunately, you don't need to perform these calculations manually. The K-Centroids Cluster Analysis tool has a simple configuration option for this.

As the text confirms, and as we saw in the image earlier (LINK), the tool provides a checkbox to Standardize the fields. When you check this box, you can choose between:

• z-score (Standardization)
• unit interval (Min-Max Normalization)

For most clustering tasks you'll encounter, selecting z-score is the standard and safest choice.

Conclusion

In this lesson, we covered the critical data preparation steps that precede any clustering analysis. Getting this stage right is more than half the battle in creating meaningful and reliable clusters.

Key Takeaways:

• Field Selection: For clustering, you must use numeric fields that are relevant to the business problem you are trying to solve.
• The Need for Scaling: Feature scaling is essential to prevent variables with large ranges (e.g., income) from dominating the distance calculations and overpowering variables with smaller ranges (e.g., age).
• Scaling Methods:
  • Standardization (Z-score): Transforms data to a mean of 0 and a standard deviation of 1. It is the most common and generally recommended method for clustering.
  • Normalization (Unit Interval): Scales data to a fixed range, typically [0, 1]. It is more sensitive to outliers.
• Alteryx Implementation: Both Z-score and Unit Interval scaling can be applied with a single click inside the K-Centroids Cluster Analysis tool.

Preview of the Next Lesson:
Now that you know how to choose your fields and prepare them by scaling, you are ready to build your first clustering model. In the next lesson, we will put this knowledge into practice and use the K-Centroids Cluster Analysis tool to build a K-Means clustering model.