Linear Discriminant Analysis Made Simple & How To Tutorial In Python

by | Nov 7, 2023 | Data Science

What is Linear Discriminant Analysis (LDA)?

Linear Discriminant Analysis (LDA) is a powerful technique in machine learning and statistics. It is primarily used for dimensionality reduction and classification tasks. LDA is a supervised learning method that finds a linear combination of features that best separates or discriminates between classes in a dataset. It is particularly valuable when you have labelled data and want to uncover the most informative features for classifying or differentiating observations.

Definition of Linear Discriminant Analysis (LDA):

  • LDA is a method that transforms high-dimensional data into a lower-dimensional space while maximizing the separation between classes.
  • Unlike Principal Component Analysis (PCA), which focuses on preserving variance, LDA aims to maximize class differences.
  • LDA is widely used in various fields, including pattern recognition, image processing, and bioinformatics, where classifying data into distinct categories is crucial.

Brief History and Context

  • R.A. Fisher introduced LDA in the 1930s as a statistical method for solving classification problems.
  • It has since evolved and found applications in diverse areas, driven by the growth of data-driven decision-making and machine learning.
  • The method has gained renewed interest with the rise of deep learning and neural networks, as it provides an effective way to reduce the dimensionality of data before feeding it into complex models.

Importance in Data Analysis and Machine Learning:

  • LDA is a fundamental tool in data preprocessing and feature engineering, as it helps in selecting the most relevant features and reducing the noise in high-dimensional datasets.
  • It is a popular choice in classification problems, such as spam email detection, facial recognition, and disease diagnosis.
  • Understanding LDA is essential for data scientists, machine learning engineers, and researchers who deal with complex datasets and wish to improve the performance of their models.

In the following sections of this blog post, we will delve deeper into the principles and applications of Linear Discriminant Analysis, exploring how it works and when to use it in various real-world scenarios.

Understanding the Basics of Linear Discriminant Analysis (LDA)

Linear Discriminant Analysis (LDA) is a valuable technique for dimensionality reduction and classification, but to grasp its full potential, it’s essential to understand its fundamental principles and how it differs from other methods, such as Principal Component Analysis (PCA).

The Problem of Dimensionality

In many real-world datasets, the number of features (dimensions) can be large. High dimensionality can lead to several challenges, including increased computational complexity, overfitting, and difficulty in visualization.

LDA aims to address the problem of dimensionality by transforming the data into a lower-dimensional space while preserving the information necessary for classification or discrimination.

Linear Discriminant Analysis (LDA) vs. Principal Component Analysis (PCA)

  • LDA and PCA are both dimensionality reduction techniques, but they serve different purposes.
  • PCA focuses on preserving variance and is unsupervised. It creates linear combinations of features that capture the maximum variance in the data.
  • LDA, on the other hand, is supervised and concentrates on maximizing the separation between classes. It finds linear combinations of features that best discriminate between different classes.
  • LDA is beneficial when the primary goal is classification or when class information is available, while PCA is more suitable for exploratory data analysis and reducing noise.

Key Assumptions of Linear Discriminant Analysis (LDA)

  • LDA relies on several assumptions to be effective:
  1. Data should be normally distributed within each class.
  2. The classes should have similar covariance matrices.
  3. Features should be independent and have equal variance.
  • Violations of these assumptions can impact the performance of LDA, so it’s essential to assess their validity in your specific dataset.

In the subsequent sections, we will explore how LDA works in practice, including its step-by-step implementation and role in two-class and multiclass classification. Understanding these basics will provide a solid foundation for applying LDA effectively in various data analysis and machine learning tasks.

Discriminant Analysis In Machine Learning

Machine learning, or Linear Discriminant Analysis (LDA) or simply Discriminant Analysis, is a classification and dimensionality reduction technique that separates data into distinct categories or classes. It is a supervised learning method widely employed in various fields, including statistics, pattern recognition, and machine learning. Discriminant Analysis aims to find linear combinations of features that maximize the separation between classes while minimizing the variance within each class.

Here are vital aspects of Discriminant Analysis in machine learning:

  • Classification: One of the primary applications of Discriminant Analysis is in classification tasks. Given a dataset with labelled classes, Discriminant Analysis finds linear discriminant functions that can separate data into these classes. When a new data point is encountered, these functions can be used to assign it to one of the classes.
  • Dimensionality Reduction: Discriminant Analysis also provides dimensionality reduction capabilities. It reduces the number of features while preserving the most relevant information for classification. This is achieved by projecting data into a lower-dimensional space where the separation between classes is maximized.
  • Two-Class and Multiclass Discriminant Analysis: Discriminant Analysis can be used for two and multiclass classification problems. In two-class Discriminant Analysis, the goal is to separate data into two classes. For multiclass problems, it extends to handle more than two classes, creating discriminant functions for each class.
  • Assumptions: Discriminant Analysis assumes that the data follows a multivariate normal distribution within each class and that the classes have similar covariance matrices. These assumptions are essential for the method to work effectively.
  • Comparison with Logistic Regression: Discriminant Analysis is often compared to Logistic Regression as they both are used for classification. However, they have different underlying principles. LDA seeks to maximize the separation between classes, while Logistic Regression models the probability of class membership directly.
  • Regularized and Non-linear Variants: There are variations of Discriminant Analysis, such as Regularized Discriminant Analysis (RDA) and Quadratic Discriminant Analysis (QDA), which provide more flexibility in handling specific data characteristics and can reduce the risk of overfitting.
  • Real-World Applications: Discriminant Analysis is used in various domains, including image recognition, speech recognition, medical diagnosis, text classification, and finance. It is precious when interpretability and class separation are essential.
  • Code Implementation: Many machine learning libraries, such as scikit-learn in Python, implement Discriminant Analysis. These libraries make it easy to apply Discriminant Analysis in practical machine learning projects.

In summary, Discriminant Analysis is a versatile technique in machine learning, offering both classification and dimensionality reduction capabilities. By understanding its principles and assumptions, you can effectively apply it to solve a wide range of real-world problems, mainly when dealing with labelled data and the need for feature selection or reduction.

Linear Discriminant Analysis (LDA) in Two-Class Classification

Linear Discriminant Analysis (LDA) is an effective tool for binary classification problems where the goal is to separate data into two distinct classes. In this section, we will delve into the specifics of applying LDA to two-class classification tasks.

Formulation of the Linear Discriminant Analysis (LDA) Algorithm

LDA begins with a straightforward formulation for two-class classification. It calculates the optimal linear discriminant by maximizing the between-class variance relative to the within-class variance.

Mathematically, LDA seeks to find the linear transformation that maximizes Fisher’s Discriminant Ratio, which measures how well the data points of different classes are separated.

Linear Discriminant Analysis (LDA) visual explanation

Example of LDA in two class classification

Step-by-Step Explanation of the Linear Discriminant Analysis (LDA) Process

  1. Calculate the mean vectors for each class in the dataset.
  2. Compute the scatter matrices: within-class scatter matrix (SW) and between-class scatter matrix (SB).
  3. Solve the eigenvalue problem for SW^(-1) * SB to obtain the discriminant vectors.
  4. Sort the discriminant vectors by their corresponding eigenvalues in descending order.
  5. Choose the top discriminant vectors as the transformation for dimensionality reduction.
  6. Project the data into the new subspace formed by the selected discriminant vectors.
  7. Perform classification using a simple decision boundary (e.g., a threshold).

Visualization of Linear Discriminant Analysis (LDA) in 2D

To better understand LDA in two-class classification, it’s helpful to visualize how the data is transformed into a lower-dimensional space.

LDA can reduce the dimensionality from the original feature space to a one-dimensional space (in the case of two-class classification) where the two classes are well-separated.

By applying LDA in two-class classification scenarios, you can effectively reduce the dimensionality of your data while preserving the class-discriminative information. This results in improved classification performance and enhanced interpretability. In the following sections, we will explore how LDA can be extended to handle multiclass classification and its practical applications in real-world scenarios.

Multiclass Linear Discriminant Analysis (LDA)

Linear Discriminant Analysis (LDA) is not limited to two-class classification problems; it can be extended to handle scenarios with more than two classes. In this section, we will explore how LDA is adapted for multiclass classification tasks.

Extending LDA to Handle More than Two Classes

  • In multiclass LDA, the objective is to find a linear transformation of the data that maximizes the separation between multiple classes.
  • Unlike in two-class LDA, where you have one discriminant function, in the multiclass case, you can have multiple discriminant functions, one for each class.
  • Each discriminant function aims to maximize the between-class variance while minimizing the within-class variance, similar to the two-class case.

The Role of Discriminant Functions

  • In multiclass LDA, each discriminant function captures the class-specific information and helps distinguish one class from the others.
  • The number of discriminant functions in multiclass LDA is typically limited to the number of classes minus one.
  • The projected data points are then compared using these functions to determine the class membership.

Example Applications of Multiclass LDA

Multiclass LDA is employed in a variety of fields and applications, including:

  1. Handwriting recognition: Recognizing handwritten characters or digits.
  2. Speech recognition: Identifying spoken words or phrases from audio data.
  3. Medical diagnostics: Classifying diseases based on medical test results.
  4. Document categorization: Sorting documents into multiple categories based on their content.

Challenges and Considerations

  • Multiclass LDA introduces complexities compared to its two-class counterpart. It involves solving a generalized eigenvalue problem, and choosing the number of discriminant functions becomes critical.
  • Data imbalance among classes, non-linear separability, or high dimensionality can present challenges that require careful preprocessing and modelling choices.

Understanding how to extend LDA to multiclass classification tasks is essential for tackling real-world problems where data may belong to multiple categories. In the subsequent sections, we will compare LDA to other classification methods, explore practical tips and best practices, and provide real-world case studies to illustrate the versatility and power of Linear Discriminant Analysis.

Linear Discriminant Analysis (LDA) vs. Logistic Regression

Regarding classification tasks in machine learning, Linear Discriminant Analysis (LDA) and Logistic Regression are commonly used methods. In this section, we’ll compare and contrast LDA with Logistic Regression, highlighting their strengths and weaknesses in various scenarios.

Comparison of LDA and Logistic Regression for Classification:

LDA and Logistic Regression are supervised learning techniques used for classification tasks, but they have distinct characteristics.


  • Both methods are used for binary and multiclass classification.
  • They are interpretable models, allowing you to understand the influence of input features on the predicted outcomes.
  • LDA and Logistic Regression can handle linearly separable data and can be extended to non-linear data through feature engineering or kernel tricks.


  • Data Requirement:
    • LDA assumes that the data is normally distributed within each class and that the classes have similar covariance matrices. Logistic Regression makes no such assumptions about data distribution.
  • Dimensionality Reduction:
    • LDA provides dimensionality reduction as a natural byproduct, while Logistic Regression does not inherently perform dimensionality reduction.
  • Linear Separability:
    • LDA explicitly seeks linear combinations of features to maximize class separability, making it well-suited for data that exhibits clear class separation. Logistic Regression, on the other hand, focuses on modelling the probability of class membership directly.
  • Model Output:
    • LDA provides class-specific discriminant scores and allows you to set custom decision boundaries, which can be advantageous in some cases. Logistic Regression provides class probabilities.
  • Overfitting:
    • Logistic Regression can be more prone to overfitting if the feature space is high-dimensional or the dataset is small, as it doesn’t inherently address dimensionality reduction like LDA.
  • Applicability:
    • LDA is often preferred when the primary goal is dimensionality reduction and maximizing class separation. Logistic Regression is versatile and widely used for various classification problems, especially when interpretability is essential.

When to Choose LDA Over Logistic Regression:

Choose LDA when:

  1. Data is normally distributed within classes and has similar covariance matrices.
  2. Dimensionality reduction is desired as a part of the classification process.
  3. Class separation is a critical objective.

When to Choose Logistic Regression Over LDA:

Choose Logistic Regression when:

  1. Data distribution assumptions do not hold or are unknown.
  2. A simple, interpretable model is needed for prediction.
  3. Feature importance is a key concern.
  4. Class separation is not as pronounced, and the primary focus is on modelling class probabilities.

Understanding the differences and trade-offs between LDA and Logistic Regression is crucial for selecting the most appropriate method for your classification problem. In the upcoming sections, we will explore practical tips and best practices for implementing LDA and provide real-world examples to illustrate its applications.

Linear Discriminant Analysis (LDA) in Dimensionality Reduction

One of the key advantages of Linear Discriminant Analysis (LDA) is its ability to reduce dimensionality while enhancing class separability. In this section, we will delve into how LDA accomplishes dimensionality reduction and its role in feature selection.

How LDA Can Be Used for Feature Selection and Reduction:

  • LDA transforms the high-dimensional feature space into a lower-dimensional space while maximizing class separability.
  • During this transformation, LDA identifies the linear combinations of features most informative for class discrimination.
  • The reduced-dimensional space, the discriminant subspace, captures the essential information necessary for classification.

Trade-Offs in Dimensionality Reduction:

  • The dimensionality reduction achieved by LDA is typically a compromise between retaining sufficient discriminative information and reducing noise.
  • Selecting the correct number of discriminant dimensions is crucial; using too few may lead to information loss while using too many can increase the risk of overfitting.

Practical Benefits of LDA in Dimensionality Reduction:

  • Improved Classification: Reduced dimensionality often improves classification performance as the model focuses on the most relevant information.
  • Enhanced Interpretability: Lower-dimensional data is more straightforward to visualize and interpret, making it valuable for exploratory data analysis.
  • Faster Training and Inference: Reduced feature space leads to quicker model training and prediction.

Steps for Implementing LDA in Dimensionality Reduction:

  1. Standardize the data: Ensure that features have zero mean and equal variance.
  2. Compute class-specific means and scatter matrices.
  3. Solve the generalized eigenvalue problem for SW^(-1) * SB, where SW is the within-class scatter matrix, and SB is the between-class scatter matrix.
  4. Sort and select the top k eigenvectors (discriminant directions) corresponding to the k largest eigenvalues to create the transformation matrix.
  5. Project the data into the discriminant subspace using the transformation matrix.
  6. Optionally, reduce the dimensionality further by selecting a subset of the transformed dimensions.

LDA’s dimensionality reduction capabilities are advantageous when dealing with high-dimensional datasets or when you want to improve the efficiency and interpretability of your machine learning models. In the subsequent sections, we will explore practical tips, best practices, and case studies to demonstrate the application of LDA in real-world scenarios.

Implementing Linear Discriminant Analysis (LDA) in Python (Code Example)

To implement Linear Discriminant Analysis (LDA) in Python, you can use the scikit-learn library. Below is a simple code example that demonstrates how to perform LDA using scikit-learn:

# Import necessary libraries
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from sklearn.datasets import load_iris
import matplotlib.pyplot as plt

# Load the Iris dataset as an example
data = load_iris()
X =
y =
target_names = data.target_names

# Create an LDA model
lda = LinearDiscriminantAnalysis(n_components=2)  # You can specify the number of components for dimensionality reduction

# Fit the model to the data
X_lda =, y).transform(X)

# Visualize the results
colors = ['navy', 'turquoise', 'darkorange']
lw = 2

for color, i, target_name in zip(colors, [0, 1, 2], target_names):
    plt.scatter(X_lda[y == i, 0], X_lda[y == i, 1], alpha=.8, color=color,

plt.legend(loc='best', shadow=False, scatterpoints=1)
plt.title('LDA of IRIS dataset')
Linear Discriminant Analysis visualizing the results by creating a scatter plot showing reduced data points in the LDA space, where colours distinguish different classes.

In this code example, we first import the necessary libraries, including scikit-learn. We load the Iris dataset as a sample dataset for illustration. You can replace this dataset with your data.

Next, we create an LDA model using the LinearDiscriminantAnalysis class from scikit-learn. You can specify the number of components you want to reduce your data using the n_components parameter.

We fit the LDA model to our dataset using the fit method and then transform the data into the lower-dimensional space using the transform method.

Finally, we visualize the results by creating a scatter plot showing reduced data points in the LDA space, where colours distinguish different classes.

This code example demonstrates how to apply LDA to a dataset in Python using scikit-learn, but you can adapt it to your specific data and classification problems.


In this comprehensive exploration of Linear Discriminant Analysis (LDA), we have delved into the fundamental principles, practical applications, and best practices of this powerful machine learning technique. LDA offers a valuable toolset for classification and dimensionality reduction, making it a versatile and widely used method in various fields.

Throughout this blog post, we’ve covered the following key points:

  • Understanding the Basics of LDA: We explained the essential concepts, including the challenges of high dimensionality, the differences between LDA and Principal Component Analysis (PCA), and the key assumptions underlying LDA.
  • Discriminant Analysis in Machine Learning: We introduced the broader concept of Discriminant Analysis in machine learning, emphasizing its significance in classification and dimensionality reduction.
  • LDA in Two-Class Classification: We detailed the formulation of the LDA algorithm for two-class classification, offering a step-by-step guide to how LDA works and its visualization in a 2D space.
  • Multiclass LDA: We extended LDA to handle multiclass classification, explaining the role of discriminant functions and providing examples of real-world applications.
  • LDA vs. Logistic Regression: We compared LDA with Logistic Regression, highlighting their similarities, differences, and the scenarios in which one is preferable.
  • LDA in Dimensionality Reduction: We explored how LDA can effectively reduce dimensionality while enhancing class separability, providing tips for implementation and highlighting its benefits.
  • Practical Tips and Best Practices: We discussed essential considerations, such as data preprocessing, avoidance of overfitting, adherence to data distribution assumptions, and feature selection, which are crucial for successful LDA implementation.
  • Implementing LDA in Python: We provided a code example using scikit-learn to demonstrate how to apply LDA to a dataset for dimensionality reduction and visualization.

Linear Discriminant Analysis, focusing on maximizing class separability and reducing dimensionality, equips data scientists, researchers, and machine learning practitioners with a powerful tool for tackling complex classification problems. Understanding when and how to apply LDA, along with adhering to best practices, allows for more accurate and interpretable results in a wide range of real-world applications.

As you continue your journey in machine learning, remember that LDA is a valuable addition to your toolkit, offering opportunities for improving the efficiency and effectiveness of your classification models and data analysis endeavours.

About the Author

Neri Van Otten

Neri Van Otten

Neri Van Otten is the founder of Spot Intelligence, a machine learning engineer with over 12 years of experience specialising in Natural Language Processing (NLP) and deep learning innovation. Dedicated to making your projects succeed.

Recent Articles

online machine learning process

Online Machine Learning Explained & How To Build A Powerful Adaptive Model

What is Online Machine Learning? Online machine learning, also known as incremental or streaming learning, is a type of machine learning in which models are updated...

data drift in machine learning over time

Data Drift In Machine Learning Explained: How To Detect & Mitigate It

What is Data Drift Machine Learning? In machine learning, the accuracy and effectiveness of models heavily rely on the quality and consistency of the data on which they...

precision and recall explained

Classification Metrics In Machine Learning Explained & How To Tutorial In Python

What are Classification Metrics in Machine Learning? In machine learning, classification tasks are omnipresent. From spam detection in emails to medical diagnosis and...

example of a co-occurance matrix for NLP

Co-occurrence Matrices Explained: How To Use Them In NLP, Computer Vision & Recommendation Systems [6 Tools]

What are Co-occurrence Matrices? Co-occurrence matrices serve as a fundamental tool across various disciplines, unveiling intricate statistical relationships hidden...

use cases of query understanding

Query Understanding In NLP Simplified & How It Works [5 Techniques]

What is Query Understanding? Understanding user queries lies at the heart of efficient communication between humans and machines in the vast digital information and...

distributional semantics example

Distributional Semantics Simplified & 7 Techniques [How To Understand Language]

What is Distributional Semantics? Understanding the meaning of words has always been a fundamental challenge in natural language processing (NLP). How do we decipher...

4 common regression metrics

10 Regression Metrics For Machine Learning & Practical How To Guide

What are Evaluation Metrics for Regression Models? Regression analysis is a fundamental tool in statistics and machine learning used to model the relationship between a...

find the right document

Natural Language Search Explained [10 Powerful Tools & How To Tutorial In Python]

What is Natural Language Search? Natural language search refers to the capability of search engines and other information retrieval systems to understand and interpret...

the difference between bagging, boosting and stacking

Bagging, Boosting & Stacking Made Simple [3 How To Tutorials In Python]

What is Bagging, Boosting and Stacking? Bagging, boosting and stacking represent three distinct ensemble learning techniques used to enhance the performance of machine...


Submit a Comment

Your email address will not be published. Required fields are marked *

nlp trends

2024 NLP Expert Trend Predictions

Get a FREE PDF with expert predictions for 2024. How will natural language processing (NLP) impact businesses? What can we expect from the state-of-the-art models?

Find out this and more by subscribing* to our NLP newsletter.

You have Successfully Subscribed!