ClusterMind

Empower Your Data. Discover New Segments. Unlock Insights.

Comment

Inspiration

This idea is taken from my scientific paper, namely in my thesis with the title "THE SEGMENTATION OF NEW STUDENT CANDIDATES USING CLUSTERING ALGORITHMA FOR PROMOTIONAL STRATEGY RECOMMENDATION OF THE NEW STUDENT ADMISSION". Where I used clustering in conducting experiments and I used Density Base K-Means and K-Medoids, after getting the results I planned to make the application but did not think what the results would be, then there was this “World's Largest Hackathon presented by Bolt” event so I could develop my thesis in the form of an application.

What it does

In the ClusterMind application, users can upload datasets from xls, xlsx and csv files, then preview the data then check whether the data is numbered or still categorical, if it is still categorical it will be made numbered. after the numbered data is obtained, proceed to normalization with the Z Transformation method, then clustering is done by choosing what method, currently only available K-Means and K-Medoids then determine the number of k values or clusters, we provide a minimum of 5 for experimentation and can be added if needed. After all the results are obtained, the value of the cluster results is Avg. within centroid distance and Davies Bouldin Index after that an evaluation is carried out using the Elbow Method until the optimal cluster results are found. after getting it, segmentation is carried out and insights from the dataset that has been uploaded.

How we built it

We built the ClusterMind application with a basic foundation of React and TypeScript coupled with Tailwind CSS, we also used ml-kmeans and ml-kmedoids to do clustering.

Challenges we ran into

There were many obstacles in building the ClusterMind application, including the following:

  1. There are many bugs that are not in line when clustering is done
  2. Bolt's lack of understanding of clustering
  3. There are many errors that I don't understand, but can be solved by bolt
  4. And many others, that I can't mention.

Accomplishments that we're proud of

I am very proud of bolt for being able to customize the application it made to what I wanted, especially during the clustering process, elbow method, cluster segmentation and insight search in the dataset. this process is very complicated and difficult for me to do myself, with the help of bolt I can do this process. Although there are still many bugs but it can be resolved perfectly to exceed my expectations.

What we learned

In the process, the things we learned were how to use clustering, how to determine effective clustering, how to determine segmentation from cluster results, how to determine insights from segmentation results and implement them.

What's next for ClusterMind

In the future we will try to add some clustering algorithms, due to limitations we can only facilitate 2 algorithms (K-Means and K-Medoids) for now. If it is possible to continue this project, various algorithms for clustering will be added and also for determining the right cluster can be added in addition to the Elbow method and for the value of the cluster results not only Avg. within centroid distance and Davies Bouldin Index.

Share this project:

Updates