Kernel-induced Online Agglomerative Clustering

I developed a Machine Learning algorithm called KOAC or Kernel-induced Online Agglomerative Clustering

1. Kernel

• The general idea is to increase the computational power of traditional linear Machine Learning algos by mapping the data into a high-dimensional feature space.
• This technique is usually refered to as the “kernel method” in ML theory.
• Inspired from the paper: “Improving the robustness of online agglomerative clustering method based on kernel-induce distance measures”. By Daoqiang Zhang, Songcan Chen, Keren Tan.

2. Online

• Just the same principle as an online K-means in N dimensions.
• Stream the data. For each new data point:
  1. Find the closest cluster
  2. Assign the new data point to this cluser
  3. Update the cluster centroid accordingly: new_centroid = old_centroid + (new_datapoint - old_centroid) / cluster_size

3. Agglomerative Clustering

• Inspired from the paper: “An on-line agglomerative clustering method for non-stationary data”. By I. D. Guedalia, M. London, M. Werman.

You can find the algo on my GitHub along with other ML projects.

---