Python Programming for Biology: Bioinformatics and Beyond

Figure 23.8.  Results from linear discriminant analysis of 2D data to separate two

data classes. Linear discriminant analysis is a method to find the linear combination of

features, and hence projection, that best separates different data classes. Here the

separation is measured as the ratio of the separation between class means to the combined

covariance. This separation is maximised by a line that goes through the sample means of

the two data classes, and a class discriminating point on this line (technically a hyperplane

in the original dimensions) can be estimated from the average of the two class means.

1

Ester, M., Kriegel, H.-P., Sander, J., and Xu, X. (1996). A density-based algorithm for

discovering  clusters  in  large  spatial  databases  with  noise.  In  E.  Simoudis,  J.  Han,  and

U.M.  Fayyad  (eds.),  Proceedings  of  the  Second  International  Conference  on  Knowledge

Discovery and Data Mining (KDD-96). AAAI Press. pp. 226–231.

2

Ankerst,  M.,  Breunig,  M.M.,  Kriegel,  H.-P.,  and  Sander,  J.  (1999).  OPTICS:  ordering

points  to  identify  the  clustering  structure.  ACM  SIGMOD  International  Conference  on

Management of Data. ACM Press. pp. 49–60.