Team Mate Building

Quick View

Goal: Develope a way to recommend team mates for future event, data based on applicants’ questionarie answer. (e.g. School, location, major…etc)

Metrics: Nope, this was an unsupervised learning project.

Results:

Kmodes is a clustering method focus on categorical data.
Some columns contain too many values, so appropriate clean techniques were applied to simplified data into categorical type.
The estimation is conducted by comparing cost value between variant number of clusters, the optimal point should see decreasing rate of cost started to slow.
The results shows that 5 clusters is optimal.
To maintain certain diversity, I give a small probability to let small group (e.g. ppl in Art or business major) be able to join with major group(e.g. ppl in CS major).
Of course cosine similarity is a solution to this.