The PCA : Principal Component Analysis
Correlation matrix, eigenvalues to find to find the value of each feature corresponding to feature vector, i.e., a main composition composed of formula:
Each result points to a formula y , find one of these lines y separately. There . 11 variables alone . 11 new coordinate axis by a straight line from point to distinguish.
Information must focus on the first few principal components. For example, PC1 represents 3 were variable.
Principal Component Analysis is provided the raw data can not be different from x point to the same Y .
Principal component analysis interaction can not be used on behalf of a group of factors.
PCA is a type of factor analysis, when the feature value extraction different algorithms can be selected.
Taking the first and second columns of the main component, two-dimensional map can be obtained:
By changing the axes may be differentially expressed more clearly.
PCA and cluster difference is Cluster goal is to y classified, PCA will feature value classification.
Correspondence Analysis: Chi-square analysis differences reflect expectations and observations, that is, information points, irregularities among the ranks. Call the shots for the chi-square matrix component analysis, the original matrix and the transposed matrix have to do it again. and so
PCA and CA comparison:
CA requires the raw data may not monotonous, do not require the normality.
PCA requires the raw data may not monotonous, the main component into the final one Euclidean distance, normal requirements.