Linear Discriminant Analysis (LDA) Flashcards
Linear Discriminant Analysis (LDA)
Linear Discriminant Analysis (LDA) is a method used in statistics, pattern recognition, and machine learning to find a linear combination of features that separates or characterizes two or more classes of objects or events. In summary, LDA is a robust and widely-used method for dimensionality reduction and classification that can be particularly effective when the assumptions of LDA are met.
- Definition
Linear Discriminant Analysis (LDA) is a method that seeks to find a linear combination of features that characterizes or separates two or more classes of objects or events. The resulting combination may be used as a linear classifier, or more commonly, for dimensionality reduction before classification.
- Goal
The primary goal of LDA is to project a feature space (a dataset n-dimensional samples) onto a smaller subspace k (where k ≤ n-1) while maintaining the class-discriminatory information.
- Procedure
LDA starts by computing the separability between different classes and within each class. The within-class scatter is computed as the sum of the scatter matrices for each individual class. The between-class scatter is computed as the scatter between the mean of each class and the overall mean. The goal is to find a vector that maximizes the between-class scatter and minimizes the within-class scatter.
- Assumptions
LDA assumes that the observations within each class are drawn from a multivariate Gaussian distribution with a class-specific mean vector and a covariance matrix that is common to all K classes.
- Multiclass LDA
While originally developed for binary classification problems, LDA can be extended to handle multiclass problems. This is done by transforming the multiclass problem into multiple binary classification problems.
- Benefits
LDA can be relatively simple to understand and implement, especially compared to some other dimensionality reduction techniques. It is a good technique to use when classes are well-separated, and the dataset is small compared to the number of features. It also performs well when the distribution of predictors is approximately normal.
- Limitations
If the assumptions of LDA (multivariate normality and equal covariance matrices) are not met, the method can produce inaccurate and misleading results. LDA can also be sensitive to outliers.
- Comparison to PCA
Unlike Principal Component Analysis (PCA), which encodes the data in a way that the total variance of the transformed data is maximized, LDA encodes the data in a way that maximizes the separability among the known categories (classes).
- Applications
LDA is used widely in pattern recognition, including computer vision and image recognition tasks. It’s also used in machine learning and statistics, especially in the pre-processing step for pattern classification and machine learning applications.