This paper presents three approaches to make the naive Bayes classifier discrimination-free. The goal is to ensure that the classifier's predictions are independent of a sensitive attribute, such as gender or race, while maintaining predictive accuracy. The first approach modifies the probability distribution of the sensitive attribute values given the class values. The second approach trains separate models for each sensitive attribute value and balances them. The third approach introduces a latent variable to represent an unbiased label and uses expectation maximization to optimize the model parameters. The paper includes experiments on both artificial and real-world data to evaluate the effectiveness of these approaches. The results show that the second approach, which trains separate models for each sensitive attribute value, performs best in terms of accuracy and discrimination-free classification. The paper also discusses the challenges of removing discrimination in machine learning, including the red-lining effect, where correlated attributes may indirectly lead to discrimination. The study highlights the importance of developing classifiers that are fair and unbiased, especially in contexts where discrimination is legally or ethically unacceptable.This paper presents three approaches to make the naive Bayes classifier discrimination-free. The goal is to ensure that the classifier's predictions are independent of a sensitive attribute, such as gender or race, while maintaining predictive accuracy. The first approach modifies the probability distribution of the sensitive attribute values given the class values. The second approach trains separate models for each sensitive attribute value and balances them. The third approach introduces a latent variable to represent an unbiased label and uses expectation maximization to optimize the model parameters. The paper includes experiments on both artificial and real-world data to evaluate the effectiveness of these approaches. The results show that the second approach, which trains separate models for each sensitive attribute value, performs best in terms of accuracy and discrimination-free classification. The paper also discusses the challenges of removing discrimination in machine learning, including the red-lining effect, where correlated attributes may indirectly lead to discrimination. The study highlights the importance of developing classifiers that are fair and unbiased, especially in contexts where discrimination is legally or ethically unacceptable.