Box-cox model can be simply defined as a statistical method called the Box-Cox transformation which helps in converting non-normal data into a normal distribution. 

The errors, also known as residuals, are assumed to have a normal distribution by many statistical techniques. Applying the Box-Cox transformation to our data lets us to make it more "normal" when it isn't normally distributed. This allows us to perform more tests and increase prediction accuracy. 

This is how it operates:

Functions: The data's scale is adjusted by the Box-Cox transformation by elevating it to a power (an exponent). The symbol for this exponent is λ (lambda). 
Range of λ: λ has a range of values from -5 to 5. 
Optimal Value: We experiment with several values of λ and select the value that best approximates a normal distribution curve for our set of data. 
Only Positive Information For positive data alone, the Box-Cox transformation is effective. For negative numbers, there is also a modified formula, though. 
Common Transformations:

• For positive data:
  • If λ = 0, the transformation is the natural logarithm (log).
  • If λ = 1, no transformation occurs (data remains unchanged).
  • Other values of λ lead to different transformations.
• For negative data:
  • There’s a separate formula to handle negative values.

Do not forget to use a normal probability plot to verify your data once the transformation has been applied. You're following the correct path if it appears more natural !