AWS Certified Machine Learning – SpecialtyExploratory Data AnalysisMedium
A machine learning engineer is analyzing sensor data from an industrial machine. The 'Temperature' readings are consistently positive but exhibit a highly skewed distribution with a long tail towards higher temperatures, indicating occasional overheating events. The engineer wants to transform this data to achieve a more symmetric, Gaussian-like distribution to meet the assumptions of certain statistical models. Which transformation is most appropriate for this type of data?
- APower transformation (e.g., Box-Cox or Yeo-Johnson)
- BStandardization (Z-score normalization)
- CMin-Max scaling
- DLogarithmic transformation
Show answer & explanationAnswer & explanation
Correct answer: A. Power transformation (e.g., Box-Cox or Yeo-Johnson)
While logarithmic transformation can help with right-skewed data, power transformations like Box-Cox or Yeo-Johnson are more general and can often achieve a more Gaussian-like distribution by finding the optimal transformation parameter (lambda) for the specific data. Box-Cox requires positive data, which 'Temperature' is, but Yeo-Johnson can handle zero or negative values if needed.
Why the other options are wrong
- B. Standardization scales data to zero mean and unit variance but does not change distribution shape.
- C. Min-Max scaling transforms data to a specific range but does not address skewness or distribution shape.
- D. Logarithmic transformation is often effective for right-skewed data but might not be optimal compared to power transforms.
Power Transformation
A family of transformations (e.g., Box-Cox, Yeo-Johnson) used to transform non-normally distributed data to a more Gaussian or symmetric distribution.
- Aims to stabilize variance and normalize data.
- Box-Cox requires strictly positive data.
- Yeo-Johnson can handle positive, negative, or zero values.
- Finds an optimal lambda (λ) parameter for the transformation.
Memory trick: Power transforms pinpoint the perfect push to normality.