CCNA Data Analysis Questions

R² represents the proportion of variance in the dependent variable explained by the independent variable. 0.85 means 85% is explained.

Logistic regression is used for binary classification problems, outputting probabilities.

K-means iteratively assigns each point to the nearest centroid, then recalculates centroids as the mean of points in the cluster.

The exhibit shows multiple base models (Model 1, Model 2, Model 3) trained in parallel on bootstrap samples of the data, and their predictions are combined via averaging (regression) or majority voting (classification). This parallel training with resampled data and equal-weight aggregation is the defining characteristic of bagging (Bootstrap Aggregating).

The query execution plan reveals how the database engine processes joins and filters. By checking the plan, the analyst can identify the most selective filter and rearrange the join order to reduce the number of rows processed early, which is the most impactful first step. Optimizing join order leverages existing indexes without requiring schema changes or hardware upgrades.

ANOVA compares means across three or more groups.

Accuracy, completeness, and timeliness are standard data quality dimensions.

Accuracy refers to how well data reflects reality.

Seasonality refers to regular patterns that repeat at fixed intervals, such as yearly.

Capping (winsorizing) and removal are common outlier treatments. Mean imputation is for missing values, and min-max normalization is scaling.

When the mean is greater than the median, the distribution is right-skewed (positively skewed) because the mean is pulled towards the higher values by outliers or a long right tail.

Mean imputation reduces variance and can distort relationships, especially if data is skewed. It may also bias estimates if missingness is not random.

Replacing missing values with the median is a form of mean/median/mode imputation.

A common rule is to consider any data point below Q1 - 1.5*IQR or above Q3 + 1.5*IQR as an outlier.

Independence of errors is a core assumption of linear regression, meaning the residuals (errors) should not be correlated with each other. This is critical for valid inference because correlated errors violate the Gauss-Markov theorem, leading to biased standard errors and unreliable hypothesis tests. In time series data, this assumption is often violated due to autocorrelation, which can be detected using the Durbin-Watson test.

Linear regression is a supervised learning algorithm because it learns a mapping from input features to a continuous target variable using labeled training data. The model minimizes the difference between predicted and actual values (e.g., via ordinary least squares) to make predictions on new data.

r = +0.92 indicates a strong positive linear relationship.

The mode is the measure of central tendency that identifies the most frequently occurring value in a dataset. Unlike the mean or median, the mode directly counts the frequency of each distinct value and returns the value with the highest count, making it the correct choice for this specific requirement.

Since p-value < α, we reject the null hypothesis of independence, meaning there is a significant association between satisfaction and product type.

K-means clustering is an unsupervised learning algorithm that partitions data into K distinct clusters based on feature similarity, making it ideal for segmenting customers by purchasing behavior without predefined labels. It groups customers who exhibit similar purchasing patterns, enabling the company to identify natural segments for targeted marketing.

The p-value > α, so fail to reject the null; the difference is not statistically significant. However, the observed lift is 0.5% (absolute). Sample size might be insufficient; statistical power could be low.

Distance-based algorithms like KNN require features on similar scales; min-max normalization is appropriate.

Accuracy and consistency are recognized dimensions of data quality in CompTIA Data+.

A one-sample z-test for proportions compares a sample proportion to a hypothesized population proportion. Here, the null is p=0.5. A chi-square test for goodness-of-fit could also be used, but the z-test is standard for a single proportion.

Multiple linear regression is the correct technique because the analyst needs to model a continuous outcome (sales) based on two or more predictor variables: advertising spend (continuous) and season (categorical, typically encoded as dummy variables). This allows the model to capture the independent effect of each predictor on sales, which simple linear regression cannot do because it only handles one predictor.

Data cleaning includes handling missing values, outlier treatment, and deduplication.

The elbow method suggests that adding more clusters beyond k=4 yields diminishing returns, so k=4 is a suitable number of clusters.

R² represents the proportion of variance in the dependent variable explained by the independent variables. 0.85 means 85% of the variance is explained.

The snowflake schema is a dimensional modeling technique where dimension tables are normalized into multiple related tables, reducing data redundancy. This structure is commonly used in data warehouses to improve query performance and maintainability for complex analytical queries.

Ridge regularization (L2) reduces variance by adding a penalty term proportional to the square of the coefficients, which shrinks them toward zero without eliminating them. This directly addresses high variance (overfitting) by constraining the model's complexity, while low bias indicates the model fits the training data well. The regularization parameter λ controls the trade-off between bias and variance.

Logistic regression outputs probabilities; it uses the logistic function (sigmoid) to map linear combination to [0,1]. The coefficients represent log-odds changes. R² is for linear regression; pseudo-R² is used but not standard R².

The chi-square test requires expected frequencies ≥5, random sampling, and independence of observations.

The trend component represents the long-term direction of the data, which is exactly what the company wants to identify.

The instability in cluster assignments is caused by the algorithm's sensitivity to the scale of features and the random initialization of centroids. Normalizing the features ensures that each variable contributes equally to the distance calculations, while setting a fixed random seed makes the initial centroid selection deterministic, leading to reproducible results.

The GROUP BY clause is used to aggregate data by department because it groups rows that have the same values in the specified column(s), allowing aggregate functions like SUM, AVG, or COUNT to be applied per group. In SQL, without GROUP BY, aggregate functions would operate on the entire result set, not per department.

Data cleaning and exploratory data analysis (EDA) are foundational steps before any modeling or time series work. With missing weather data (10%) and incorrect driver IDs, proceeding without cleaning would introduce bias and errors. EDA will reveal patterns, correlations, and data quality issues, enabling informed decisions on imputation and feature engineering for the predictive model.

Listwise deletion (removing rows with missing values) is simple and unbiased if data is MCAR, but it reduces sample size. However, it is least likely to introduce bias among the options when MCAR holds.

The t-test assumes normality of the data, especially with small samples. Violation of normality can affect the validity.

The median is the most appropriate measure of central tendency when data contains extreme outliers, such as the very low customer satisfaction scores described. Unlike the mean, the median is resistant to skew because it depends only on the middle value(s) of the sorted dataset, not on the magnitude of extreme values. This makes it the standard choice for summarizing ordinal or skewed interval/ratio data in data analysis.

Capping limits extreme values to a threshold, and removal deletes outlier records. Transformation (e.g., log) can reduce impact but is more for skewness. Imputation and binning are for missing data or discretization, not directly for outliers.

Recall (also known as sensitivity or true positive rate) measures the proportion of actual positives correctly identified. With only 10% of actual churners detected, the model has a recall of 0.1, which directly highlights the failure to capture churners despite high overall accuracy.

Since p-value (0.04) < α (0.05), we reject the null hypothesis, indicating a statistically significant difference.

Cross-validation is essential for model validity because it partitions the data into multiple folds, training on k-1 folds and validating on the remaining fold, which provides a robust estimate of model performance and reduces overfitting. This technique ensures that the model generalizes well to unseen data by repeatedly testing different subsets, making it a standard practice in predictive modeling.

Handling missing values and removing duplicates are standard data cleaning tasks.

K-means clustering is the correct choice because it is an unsupervised learning algorithm that partitions data into K distinct clusters based on feature similarity. For segmenting customers by purchasing behavior (frequency, monetary value, recency), K-means groups customers with similar numeric patterns without requiring labeled outcomes, making it ideal for exploratory segmentation.

Replacing missing values with the mean (for normally distributed data) or median (for skewed data) is a standard imputation technique that preserves the central tendency of the dataset without introducing bias. This method is appropriate when the missingness is random and the proportion of missing data is low, as it maintains the sample size for analysis.

Removing exact duplicates and standardizing identifiers help resolve duplicates.

Min-max normalization rescales features to a fixed range (e.g., 0 to 1), making distances computed equally weighted. Standardization is better for algorithms assuming Gaussian distributions.

A two-sample t-test compares the means of two independent groups.

R-squared of 0.85 means 85% of the variance in house prices is explained by square footage.

Chi-square test requires categorical variables, expected frequencies >=5 in at least 80% of cells, and independence of observations.

Uniqueness refers to the expectation that each record or attribute value should be unique. Duplicate phone numbers violate uniqueness.

Since p-value (0.03) < α (0.05), we reject the null hypothesis, indicating a statistically significant difference.

Since 0.7 > 0.5, the predicted class is churn (usually coded as 1).

Classification involves defining levels, assigning ownership, labeling, access control, and training.

Pearson r ranges from -1 to 1, measuring linear relationship; +1 indicates perfect positive linear correlation.

The median is not affected by extreme values, making it robust to outliers.

ANOVA is used to compare means of three or more groups.

Min-max normalization scales features to a 0-1 range, ensuring each feature contributes equally to distance calculations.

When residuals from a regression model on time series data exhibit autocorrelation, the standard ordinary least squares (OLS) assumptions are violated, leading to biased standard errors and unreliable inference. An ARIMA model is specifically designed to handle autocorrelated time series by explicitly modeling the autoregressive (AR) and moving average (MA) components, making it the correct adjustment to capture the temporal dependencies in stock price movements.

Z-score standardization transforms data to have mean 0 and standard deviation 1, which is suitable for algorithms that assume normality (like linear regression).

Applying a standardized date parsing function (e.g., `TO_DATE` in SQL or `pd.to_datetime` in Python) ensures all date values are converted to a single, consistent format regardless of the original representation. This is a fundamental ETL best practice to avoid ambiguity and enable accurate date-based filtering, aggregation, and joins during analysis.

Mean imputation replaces missing values with the column mean.

A Q-Q plot is used to assess whether the residuals of a linear regression model are approximately normally distributed, which is a key assumption for valid inference (e.g., p-values and confidence intervals). Option E is correct because residual plots (e.g., fitted vs. residuals) are the standard diagnostic to check for homoscedasticity—constant variance of errors across all levels of the independent variables—another core assumption of ordinary least squares regression.

PCA is the correct choice because it is an unsupervised linear dimensionality reduction technique that projects the data onto orthogonal components ordered by the variance they capture. By selecting the top principal components, the analyst can retain the maximum possible variance in the dataset while reducing the number of features from 100 to a smaller set, directly addressing the goal of preserving variance.

A correlation coefficient of -0.85 indicates a strong negative linear relationship between temperature and heating costs. As temperature increases, heating costs decrease significantly, and the magnitude of 0.85 (close to -1) confirms the strength of this inverse association.

The query groups sales by product_id and filters groups where the average quantity exceeds 8. From the exhibit (not shown but implied), only product_ids P001 and P003 have an AVG(quantity) > 8, so they are returned. The HAVING clause operates on aggregated data after GROUP BY, unlike WHERE which filters rows before aggregation.

Adjusted R-squared penalizes for adding predictors that do not improve the model significantly; a gap suggests some predictors may be irrelevant or the sample size is small.

The elbow method helps choose k where adding more clusters yields diminishing returns in reducing variance.

The model is overfitting: 100% training accuracy vs. 60% test accuracy with a depth-20 tree. Increasing minimum samples per leaf forces the tree to be simpler by requiring more samples in each leaf, reducing variance and improving generalization. This directly combats the overfitting caused by the overly deep tree.

ANOVA (Analysis of Variance) is used to compare means of three or more groups.