Naive Bayes Model is based the assumption that the variables {𝑋1, …, 𝑋𝑛} areΒ conditionally independentΒ givenΒ π‘Œ

  • 𝐏(π‘Œ|𝑋1, …, 𝑋𝑛) = 𝐏(𝑋1|π‘Œ) … 𝐏(𝑋𝑛|π‘Œ) 𝐏(π‘Œ) / 𝐏(𝑋1, …, 𝑋𝑛)
  • 𝐏(π‘Œ|𝑋1, …, 𝑋𝑛) ∝ 𝐏(𝑋1|π‘Œ) … 𝐏(𝑋𝑛|π‘Œ) 𝐏(π‘Œ)

if the variablesΒ {𝑋1, …, 𝑋𝑛} happen to NOT be conditionally independent givenΒ π‘Œ, this would cause an Overcounting Problem

In other words, correlated variables in {𝑋1, …, 𝑋𝑛} which breaks conditional independence