Showing changes from revision #5 to #6:
Added | Removed | Changed
According to Wikipedia:
Principal component analysis (PCA) is a mathematical procedure that uses an orthogonal transformation to convert a set of observations of possibly correlated variables into a set of values of uncorrelated variables called principal components. The number of principal components is less than or equal to the number of original variables. This transformation is defined in such a way that the first principal component has as high a variance as possible (that is, accounts for as much of the variability in the data as possible), and each succeeding component in turn has the highest variance possible under the constraint that it be orthogonal to (uncorrelated with) the preceding components. Principal components are guaranteed to be independent only if the data set is jointly normally distributed. PCA is sensitive to the relative scaling of the original variables.
Depending on the field of application, it is also named the discrete Karhunen–Loève transform (KLT), the Hotelling transform or proper orthogonal decomposition (POD).
PCA was invented in 1901 by Karl Pearson. Now it is mostly used as a tool in exploratory data analysis and for making predictive models. PCA can be done by eigenvalue decomposition of a data covariance matrix or singular value decomposition of a data matrix, usually after mean centering the data for each attribute. The results of a PCA are usually discussed in terms of component scores (the transformed variable values corresponding to a particular case in the data) and loadings (the weight by which each standarized original variable should be multiplied to get the component score) (Shaw, 2003).
PCA with $n$ basis vectors can also be viewed as a form of “compression”/data reduction, namely the selection of an $n$ dimensional subspace $S$ such that projection of the dataset vectors to $S$ has the minimal summed $L_2$ error relative to the original dataset. (This $L_2$ minimisation interpretation is one “explanation” of the dependence of the PCA decomposition on variable scaling.)
The orthogonal transformation used by PCA is implicitly assuming that the dataset is generated by process with a zero “expected value”, so that generally a “mean vector” is subtracted from all the data points. There are two approaches to acheiving this:
Using additional knowledge about the process generating the dataset to deteremine the mean, e.g., that the mean should actually be exactly zero.
Using the computed average of the data set as the “mean vector”.
The compression viewpoint is relatively insensitive to the choice of mean, but techniques which use the vector of PCA components as random variates can be significantly affected by the mean vector.
Given an $m$-dimensional dataset $\{y^{i}\}_{i=0}^K$ the $n$ dimensional PCA is a set of $n$ -dimensional vectors orthonormal$m$-dimensional vectors $\{\mathbf{v}^j\}_{j=0}^n$ (with the “artificial” vector $\mathbf{v}^0$ often referred to as $\mu$) along with a set of $n$-dimensional “coefficient” vectors $\{\lambda^{i}\}_{i=0}^K$ which can be used to approximate represent the approximations original$y{\hat{y}}^{i}$ y \hat{y}^i s using to the original$y^i$ using
where the second expression absorbs the mean by having using a coefficient$\lambda^i_0$ which is identically always$1$ . and The crucial property is that, other than the mean vector, all the$\mathbf{v}^0 = \mu$. The crucial property is that, other than the mean vector, all the $\mathbf{v}$s are orthonormal, i.e., ${v}^{i}\cdot {v}^{j}={\delta}_{\mathrm{ij}}$ \mathbf{v}^i \cdot \mathbf{v}^j=\delta_{ij} \mathbf{v}^j=\delta_{i j} . In addition to optimising the information “reconstruction content error” for a given number of PCA basis vectors, this orthogonality gives rise to other useful calculational properties of PCA.
In the above formulation the $\lambda^i_j$s may naturally have very different magnitudes, so in some applications there is a further decomposition as
(Again the mean $\mu$ can be absorbed into an “artificial” principal component if desired.) Here the $\nu^j$ value is the “length scale” of the $\mathbf{v}^j$ principal component, so that it can generally be taken as a reasonable assumption that the loadings $w^i_j$ come from the same probability distribution, often further assumed to be the $N(0,1)$ normal distribution. (Technically the optimal $\mathbf{v}^j$ vectors turn out to be the eigenvectors of covariance matrix of the sample set, and theeigenvectors of covariance matrix of the sample set, and the $\nu^j$ s are the square roots of the corresponding eigenvalues.)eigenvalues.)