MLS-C01 Exploratory Data Analysis Practice Test 4 — 15 Questions

Question 1

Which TWO of the following are common techniques for detecting outliers in a numerical feature?

Accepted Answer

Interquartile Range (IQR). The Interquartile Range (IQR) method (C) is a standard outlier-detection technique: it computes Q1 and Q3, defines the bounds as Q1 − 1.5×IQR and Q3 + 1.5×IQR, and flags any value outside those fences as a potential outlier. The Z-score method (D) is also a common technique: it measures how many standard deviations a value lies from the mean (z = (x − μ)/σ), and values with |z| above a threshold such as 3 are treated as outliers. The Chi-square test (A) is used for categorical variables to test association or goodness-of-fit, not for finding outliers in a numerical feature. Standard deviation (B) is a dispersion measure that underlies the Z-score but is not itself an outlier-detection technique. Principal Component Analysis (E) is a dimensionality-reduction method, not a dedicated outlier-detection procedure for a numerical feature.

Answer

Chi-square test

Answer

Standard deviation

Answer

Principal Component Analysis (PCA)

Question 2

During exploratory data analysis, a data scientist observes a strong correlation (r=0.95) between two numeric features. The model to be trained is a linear regression. What is the most appropriate action?

Accepted Answer

Remove one of the correlated features.. When two numeric features are highly correlated (r=0.95), they carry nearly redundant information, which causes multicollinearity in linear regression — inflating coefficient variance and making the model unstable and hard to interpret. Removing one of the correlated features is the cleanest fix because it eliminates the redundancy without adding complexity, and the remaining feature preserves almost all the predictive signal.

Answer

Apply standardization to both features.

Answer

Use L2 regularization (Ridge regression) without removing features.

Answer

Create an interaction term between the two features.

Question 3

A data scientist is exploring a dataset with 1,000 features and only 200 samples. The goal is to build a binary classifier. Which technique should be used first during exploratory data analysis to reduce dimensionality and avoid overfitting?

Accepted Answer

Use principal component analysis (PCA) to reduce dimensions.. PCA is the correct first step because it is an unsupervised linear transformation that projects the 1,000 features onto a smaller set of orthogonal principal components ranked by explained variance, directly reducing dimensionality before any classifier is trained. With only 200 samples and 1,000 features (a p >> n regime), PCA combats the curse of dimensionality and multicollinearity, lowering variance and the risk of overfitting. It is fast, deterministic, and does not require the target labels, making it ideal for early exploratory data analysis.

Answer

Compute pairwise correlations and remove highly correlated features.

Answer

Apply L1 regularization (Lasso) to select features.

Answer

Use t-SNE to visualize clusters and reduce dimensions.

Question 4

A data scientist is performing EDA and observes that a feature 'purchase_amount' has many zeros and a long tail of positive values. What type of model would be appropriate for this target variable?

Accepted Answer

Zero-inflated negative binomial regression.. A target variable with many zeros and a long right tail of positive values is characteristic of zero-inflated count data, which is best modeled with zero-inflated negative binomial regression. This approach handles both the excess zeros (via a separate inflation component) and the overdispersed positive counts (via the negative binomial distribution).

Answer

Linear regression after log transformation.

Answer

Logistic regression on binary indicator of purchase.

Answer

Poisson regression.

Question 5

Which THREE are valid reasons to perform feature scaling during exploratory data analysis?

Accepted Answer

To improve performance of distance-based algorithms like KNN.

Answer

To change the shape of the feature distribution.

Answer

To increase the number of features.

Question 6

A data scientist is analyzing a dataset with missing values in a numeric column. The missing rate is 30% and the data is not missing completely at random. Which imputation method should the data scientist avoid to minimize bias?

Accepted Answer

Mean imputation. Mean imputation (Option A) should be avoided when data is not missing completely at random (NMAR) because it can introduce bias by underestimating variance and distorting the relationships between variables. Options B (model-based imputation), C (k-NN imputation), and D (multiple imputation) are more robust for non-random missing data as they account for patterns in the data and produce less biased estimates.

Answer

Model-based imputation using linear regression

Answer

k-Nearest Neighbors imputation

Answer

Multiple imputation using chained equations

Question 7

A machine learning engineer is exploring a dataset with 500 features and 10,000 samples. To reduce dimensionality for visualization, which technique is most suitable if the goal is to preserve global data structure?

Accepted Answer

Principal Component Analysis (PCA). PCA is the most suitable technique for preserving the global data structure when reducing dimensionality because it is a linear method that maximizes variance along orthogonal principal components, capturing the overall covariance structure of the 500 features. Unlike nonlinear methods, PCA ensures that the global relationships (e.g., distances between clusters) are retained, making it ideal for visualization of high-dimensional data where the goal is to see broad patterns.

Answer

t-Distributed Stochastic Neighbor Embedding (t-SNE)

Answer

Locally Linear Embedding (LLE)

Answer

Uniform Manifold Approximation and Projection (UMAP)

Question 8

A data analyst is working with a time series dataset that shows increasing variance over time. To stabilize the variance before modeling, which transformation is most appropriate?

Accepted Answer

Log transformation. The log transformation is specifically designed to stabilize variance when the standard deviation grows proportionally with the mean, which is exactly what happens in a time series with increasing variance (heteroscedasticity). By compressing larger values more than smaller ones, it makes the variance more constant across the series. While Box-Cox is a more general family that includes the log as a special case, the question asks for the most appropriate transformation given the pattern, and log is the canonical choice for exponential growth in variance. First-order differencing removes trend, not variance, and min-max scaling only rescales without changing the variance structure.

Answer

First-order differencing

Answer

Box-Cox transformation

Answer

Min-max scaling

Question 9

A team is analyzing a dataset with many categorical features. They notice that one feature has 1,000 unique values but a long tail where most values appear only once. Which encoding method is most appropriate to avoid overfitting?

Accepted Answer

Count encoding. Count encoding replaces each category with the number of times it appears in the dataset, which handles high-cardinality features with long tails without creating a sparse, high-dimensional matrix. Because rare categories (appearing once) get a count of 1, the model can treat them uniformly as low-frequency rather than memorizing each unique value, reducing overfitting. It also preserves a signal about frequency that can be predictive while keeping dimensionality at one column.

Answer

Target encoding

Answer

Label encoding

Answer

One-hot encoding

Question 10

A machine learning engineer is analyzing a dataset and observes that the distribution of a continuous feature is heavily right-skewed. Which transformation is most likely to make the distribution approximately normal?

Accepted Answer

Log transformation. A log transformation compresses the long right tail of a right-skewed distribution, pulling large values closer to the bulk of the data and making the distribution more symmetric and approximately normal. It is the standard first choice for positively skewed continuous features. Box-Cox with lambda=0 is mathematically equivalent to a log transform, but the question asks for the most likely transformation, and log is the direct, interpretable answer.

Answer

Square root transformation

Answer

Exponential transformation

Answer

Box-Cox transformation with lambda = 0

Question 11

Which THREE of the following are common causes of multicollinearity in a linear regression model?

Accepted Answer

Including a polynomial term (e.g., x^2) along with the original variable. Options A, C, and D are correct. Dummy variable trap occurs when all categories are included without dropping one. Highly correlated predictors directly cause multicollinearity. Including polynomial terms creates correlation with the original variable. B (interaction terms) can also cause but is less common. E (outliers) does not cause multicollinearity.

Answer

Including interaction terms between independent variables

Answer

Presence of outliers in the target variable

Question 12

A machine learning engineer is analyzing feature distributions in a dataset and notices that one feature has a long tail. Which transformation is most appropriate to reduce skewness and make the distribution more normal?

Accepted Answer

Apply a log transformation. Log transformation is the most appropriate technique to reduce right skewness (long tail) and make the distribution closer to normal. One-hot encoding is used for categorical variables, not for transforming skewed numerical features. Min-max normalization scales features to a range but does not change the shape of the distribution. Standardization (Z-score) centers the data and scales by standard deviation, but also does not reduce skewness.

Answer

Apply one-hot encoding

Answer

Apply min-max normalization

Answer

Apply standardization (Z-score)

Question 13

A machine learning team is analyzing feature importance in a dataset with many categorical features. They plan to use a tree-based model. Which encoding method should they use to handle high-cardinality categorical features without creating too many dummy variables?

Accepted Answer

Target encoding. Target encoding replaces categories with the mean of the target, which is efficient and works well with tree models. Option A is wrong because one-hot encoding creates many columns for high cardinality. Option B is wrong because label encoding imposes ordinality. Option D is wrong because frequency encoding may not capture predictive information.

Answer

One-hot encoding

Answer

Label encoding

Answer

Frequency encoding

Question 14

A data engineer is performing exploratory data analysis on a dataset with 1 million rows and 50 features. The engineer wants to identify missing values and outliers. Which THREE approaches should the engineer use? (Choose three.)

Accepted Answer

Use a DataFrame.info() method to see non-null counts. Options B, D, and E are correct because they directly help identify missing values and outliers. DataFrame.info() shows non-null counts per column, revealing missing values. The missingno matrix visualizes missing data patterns across rows and columns. DataFrame.describe() provides summary statistics (count, mean, std, min, max, quartiles) that can indicate outliers (e.g., values far outside the interquartile range). Option A (correlation heatmap) is used to assess relationships between features, not to detect missing values or outliers. Option C (box plots for all features simultaneously) is impractical with 50 features because it would be cluttered and hard to interpret; box plots are more useful when applied selectively to a few features or after filtering.

Answer

Create a correlation heatmap of all features

Answer

Plot box plots for all features simultaneously

Question 15

A data scientist is exploring a dataset with a binary target variable. Which TWO metrics are appropriate for evaluating the balance of the target classes? (Choose two.)

Accepted Answer

Count plot of the target variable. A Count plot of the target variable is correct because it directly visualizes how many observations fall into each of the two target classes, making class imbalance immediately apparent. D value_counts() on the target column is correct because it returns the exact frequency of each class label, which is the standard quantitative way to assess target balance. B Histogram of a feature describes the distribution of a predictor, not the target classes, so it does not measure class balance. C Scatter plot of two features colored by target shows feature separation and overlap between classes, not the proportion of each class. E Correlation matrix of all features measures linear relationships among variables and does not quantify the balance of the binary target.

Answer

Histogram of a feature

Answer

Scatter plot of two features colored by target

Answer

Correlation matrix of all features