3.4 Conclusion
Component 1
Rows: Vienna variants (mixed of meat)
Cols: Floury, Spicy, Rubbery, White Meat VS Dark Meat
Interpret: Component 1 is explaining “the difference within” Vienna variants
Component 2
Rows: Pavo variants (Turkey)
Cols: Rubbery & White Meat VS Acidic, Bitter, Floury, Dark Meat
Interpret: Component 1 is explaining “the difference within” Pavo variants
Performance
We see that BADA, especially in Fixed Data showed that it could classify the observations in a Supervised manner with really high accuracy (93%).
In BADA, we essentially performed PCA on the group means of the dataset. The grouping strategy that we have chosen derived from our findings in PCA regarding the differentiation between Pavo and Viena sausages. The amount of variance among group means in BADA is noticeably more pronounced than the amount of variance in PCA. One reason for this is that the variances among group means are less than the variances among all observations. One major benefit of BADA is that it can be used to predict supplementary observation (unlearned / novel data). This aspect can be very beneficial in solving many real-world prediction problem, similar to clustering methods such as K-means, H-clust, Tree, etc. In term of insights yielded, BADA largely painted a similar picture to PCA. Component 1 and 2 still do a great job at seperating the group means of the 2 sausages types Pavo vs Viena, and their spreads are still along component 1 and 2. For variables, we see similar patterns and the presence of the same set of important variables such as: Floury, Rubbery, Dark Meat, White Meat, Umami (aka HVP), Spicy, Bitter, and Acidic.