Mahalanobis distance is most useful for quantitative variables only. Presence of categorical and discrete variables along with the continuous variables make it difficult to use the Mahalanobis distance for the purpose of discriminant analysis and classification.
Hi B K Hooda,
You can use, for instance, "normalized Euclidean distance" to find the training instance closest to the given test instance. Another choice is to use "entropy-based distance function".
HTH.
Samer
Thank you Samer. But whether normalized or non-normalized Euclidean distance cannot be preferred over the Mahalanobis distance. Which is also not suitable for mixed data. However, some researchers are working on entropy based distances like Kullback–Leibler divergence but using it too we have to compromise with some of the optimum properties of a distance measure.
Hi,
Mahalanobis distance for mixed data is described in the following paper http://www.sciencedirect.com/science/article/pii/S0047259X03001507
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis