Which of the following is NOT a recommended method to handle missing data in preprocessing?

Question

BlackTom AI · Accepted Answer

Replacing missing values with 0 (zero) is not recommended in general because 0 is usually not a neutral or meaningful substitute and can introduce bias, distort distributions, and mislead models.

1) Relevant knowledge points and concepts
- Types of missingness: MCAR (missing completely at random), MAR (missing at random), MNAR (missing not at random) — choice of handling depends on this.
- Common methods: deletion (drop rows/features), simple imputation (mean/median/mode), interpolation for time series, model-based or k-NN imputation, and adding missingness indicator features.
- Risks: imputing an arbitrary constant like 0 can create artificial clusters, change feature scales, and bias estimators unless 0 is a meaningful value for that variable.

2) Step-by-step reasoning
- If 0 is not a genuine possible value (e.g., income, temperature in Celsius when negative values occur, or rating scales that start at 1), substituting 0 changes the underlying distribution and expected value.
- Many algorithms (linear models, distance-based methods) will treat the imputed zeros as real observations, altering correlations and distances.
- Better alternatives: use mean/median imputation for a quick unbiased central estimate (option d), interpolation for ordered/time data (option c), k-NN or model-based imputation to leverage relationships between features (option b), or remove rows/features when missingness is pervasive (option e).

3) Calculations
- No numeric calculation is required here; the argument is conceptual and statistical. (If one replaced missing values with mean, the sample mean remains unbiased under MCAR; replacing with 0 generally does not preserve that property unless the true mean is 0.)

4) Why the correct answer is correct
- Option (a) is not recommended because 0 is arbitrary and often not representative, causing bias and misleading results. Only use 0 when it is a legitimate, meaningful value for that feature (e.g., count of occurrences where 0 truly means none).

5) Why other options are incorrect (i.e., they are recommended)
- b: k-NN or model-based imputation is an advanced, often effective method.
- c: Linear interpolation is appropriate for time-ordered or spatial data.
- d: Column-mean imputation is a common simple strategy.
- e: Removing rows/features is appropriate when missingness is extensive or non-informative.

Conclusion: Therefore the NOT recommended method is (a) Replace missing values with 0 (zero), which matches the provided answer.

Which of the following is NOT a recommended method to handle missing data in preprocessing?单项选择题

类似问题

Some image features are missing more often for lower-quality scans. What question matters most before imputation?

There are missing data in the heart_disease dataset as shown in the table above. Provide the code to show how to handle the missing data. Explain your approach.

You own a bond that pays $64 in interest annually. The face value is $1,000 and the current market price is $1,021.61. The bond matures in 11 years. What is the yield to maturity?

Which of the following statements is CORRECT?

更多留学生实用工具

智能学习助手

风格化写作助手

论文查重助手

文献引用助手

课堂转译助手

课堂笔记助手

Quiz搜索助手

学校历年真题

智能刷题助手

智能匹配练习

希望你的学习变得更简单