Relocating negative silhouette value (or misclassified data) into neighbouring clusters

Hi everyone,

Insted of deleting a miscalssied data (negative silhouette), may I know if MATLAB provide a function or codes that can operates re-allocation of misclassified data from the silhouette analysis to a neighbouring class.

I found a code in this website https://onlinelibrary.wiley.com/doi/full/10.1111/jvs.12984 (you can see the code in supporting documents), but the code operates using the R language. Since I am not familiar or don't have any knowledge of R language. Can someone help me interpret the code into MATLAB.

Thanks.

Answer 1

[Akshat Dalal]

Hi Mat,

In MATLAB, there isn't a built-in function specifically designed for reallocating misclassified data from silhouette analysis to a neighbouring class. However, you can do it manually by creating a custom implementation. After performing Silhouette Analysis in MATLAB, you can use the following approach to reallocate Misclassified Data:

1. Reassign to the nearest neighbouring cluster: Use distance metrics, such as Euclidean distance using the ‘pdist’ function, to find the nearest neighbouring cluster for each misclassified data point. Then, reassign the data point to that neighbouring cluster.
2. Reassign based on similarity: Measure the similarity between the misclassified data point and the centroids of neighbouring clusters. Use a similarity metric, such as cosine similarity using the ‘cosineSimilarity’ function or correlation using the ‘corr’ function, to find the most similar cluster and reassign the data point accordingly.
3. Re-evaluate and Refine: After reallocating the misclassified data, re-evaluate the clustering results and assess the impact of the reallocation. You may need to iterate this process multiple times, adjusting the reallocation strategy or the clustering algorithm parameters, to achieve satisfactory results.

To read more about clustering using MATLAB, please refer the following documentation: https://www.mathworks.com/help/images/color-based-segmentation-using-k-means-clustering.html

To read more about the functions mentioned in the steps above, please refer the following documentations:

1. pdist function: https://www.mathworks.com/help/stats/pdist.html
2. cosineSimilarity function: https://www.mathworks.com/help/textanalytics/ref/cosinesimilarity.html
3. corr function: https://www.mathworks.com/help/stats/corr.html

I hope this helps.