Interactive and Robust PCA Graphics with FSDA in MATLAB
Principal Component Analysis (PCA) is a standard method for dimension reduction and the exploratory analysis of multivariate data, but a complete applied workflow often requires more than numerical eigenvectors and static plots. This article documents a set of routines in the FSDA toolbox for MATLAB that implements an integrated workflow for interactive and robust PCA graphics. The software contribution consists of coordinating classical SVD-based PCA, dynamic biplot scaling controlled by row and column parameters, score-distance and orthogonal-distance diagnostics, hard-trimming robust PCA through the Minimum Covariance Determinant and the Forward Search, linked brushing, geographical displays, and three-dimensional projection diagnostics. The implementation returns reproducible numerical output while also exposing graphical controls for changing biplot geometry, confidence regions, displayed labels, robust trimming levels, and linked selections. The methodology is illustrated on quality-of-life indicators for Italian provinces, where the software reveals variable groups, atypical provinces, robust and non-robust PCA structures, and geographical patterns. The emphasis is on statistical software: no new PCA estimator is proposed, but established PCA and robust-PCA methodology is combined in a documented, extensible, and script-reproducible MATLAB environment.