CODEWITHSUNDEEP
pythoncluster-analysisdbscan
Abboud
Abboud

Reputation: 23

DBSCAN quality check using python

I have a dataset of 781 instances where two main categories exist with numerical values. Such that there is one column called 'ΔR(mm)'. the dataset has main two categories according to the values of this column (the values with negative sign and the values of positive sign). When I run the DBSCAN algorithm in python I got 5 clusters each of mixed values (negative and positive) with varied number of instances per each cluster. E.g., 135,27, 423, 31, 145 instances in clusters 1, 2,3, 4 and 5 respectively. I set the hyperparameters eps = 1400 , and min_samples = 10 to reduce the number of noise to 11 points. So, I have two questions according to these givens:

1- Is my dataset considered as sparse? according the min_samples and eps values? I have selected the range of eps by computing the KNN then sort the distances and check the knee values enter image description here as shown in the figure. Are these hyperparameters considered reasonable?

2- How to check the quality of clustering? is it related to reducing the number of noise instances to the min?

Upvotes: 0

Views: 27

Answers (0)

Related Questions