K-Medoids clustering in Machine Learning

K-Medoids, also known as Partitioning Around Medoids (PAM), is a clustering algorithm introduced by Kaufman and Rousseeuw. It is similar to K-Means, but instead of using the mean of points as a cluster center, it uses an actual data point called a medoid.

What is a Medoid?

A medoid is the most centrally located data point within a cluster. It minimizes the total dissimilarity with all other points in that cluster. The dissimilarity between a medoid Ci and an object Pi​ is given by:

The total cost (or objective function) of K-Medoids is defined as:

K-Medoids Algorithm Steps

1. Initialize: Randomly select k data points from the dataset as initial medoids.

2. Assign Points: Assign each data point to the nearest medoid using a distance metric (e.g., Manhattan or Euclidean).

3. Update Step (Swap): For each medoid m, try swapping it with a non-medoid point ooo.

• Recalculate the cost for this new configuration.
• If the total cost decreases, accept the swap; otherwise, revert.

4. Repeat: Continue until no further cost reduction is possible.

Solved Example

Let’s consider the following example.

If a graph is drawn using the above data points, we obtain the following: 

Step 1: Initialization

Let the randomly selected 2 medoids be: k=2 and let C₁ = (4, 5) and C₂ = (8, 5).

Step 2: Calculating cost

The dissimilarity of each non-medoid point with the medoids is calculated and tabulated: 

We use the Manhattan distance formula to calculate the distance between medoid and non-medoid points:

Each point is assigned to the cluster of the medoid whose dissimilarity is lesser.

• Points 1, 2 and 5 -> Cluster C₁
• Points 0, 3, 6, 7 and 8 -> Cluster C₂

Cost = ( 3+ 4+ 4) +( 3+ 1+ 1+ 2+ 2) = 20

Step 3: Swap and Recalculate

Now, randomly select one non-medoid point and recalculate the cost. Let the randomly selected point be (8, 4).

Each point is assigned to the cluster whose dissimilarity is lesser.

• Points 1, 2 and 5 -> Cluster C₁
• Points 0, 3, 6, 7 and 8 -> Cluster C₂

New Cost = ( 3+ 4+ 4) +( 2+ 2+ 1+ 3+ 3) = 22

Swap Cost = New Cost - Previous Cost = 22-20 = 2

Since 2 >0, the swap cost is not less than zero, so we undo the swap.
Hence, (4, 5) and (8, 5) are the final medoids.

Final Result

The clustering is as follows:

• Cluster 1: (4, 5) -> Points 1, 2, 5
• Cluster 2: (8, 5) -> Points 0, 3, 6, 7, 8

The time complexity of the K-Medoids algorithm is:

Advantages

• It is simple to understand and easy to implement.
• K-Medoids converges in a fixed number of steps.
• It is less sensitive to outliers compared to other partitioning algorithms.

Disadvantages

• Not suitable for non-spherical or arbitrarily shaped clusters.
• Results may differ across runs due to random initialization of medoids.