ISYE 6501 Introduction to Analytics Modeling Final Exam Prep | 200 Questions & Answers with Rationales Academic Year: 2026/2027 | Georgia Tech OMSA Study Bundle

ISYE 6501 Introduction to Analytics Modeling
Final Exam Prep | 200 Questions & Answers
with Rationales
Academic Year: 2026/2027 | Georgia Tech
OMSA Study Bundle

SECTION 1: MODEL VALIDATION & SELECTION (Questions 1-25)



Q1: In a linear regression model, what does the p-value for a coefficient indicate?

Answer: The probability of observing the coefficient (or more extreme) if the true coefficient is
zero.

Rationale: A low p-value (<0.05) suggests the predictor is statistically significant, meaning there
is evidence that the predictor has a non-zero effect on the response variable .



Q2: Which validation method is most computationally efficient for large datasets?

Answer: Validation set approach (holdout).

Rationale: Uses a single train-test split, requiring less computation than cross-validation
methods like k-fold which must fit the model k times .



Q3: In k-nearest neighbors (k-NN), what happens if k is set to 1?

Answer: The model overfits; the decision boundary is highly irregular.

Rationale: Low k captures noise in the training data, resulting in high variance and poor
generalization to new data .



Q4: What does the coefficient of determination (R²) represent in linear regression?

,Answer: The proportion of variance in the dependent variable explained by the model.

Rationale: R² quantifies the percentage of total variation in the response variable accounted for
by predictors, ranging from 0 to 1 where higher values indicate better fit .



Q5: Which of the following is a key assumption of ordinary least squares (OLS) linear
regression?

Answer: The errors (residuals) are homoscedastic and uncorrelated.

Rationale: Key OLS assumptions include linearity, independence of errors, homoscedasticity
(constant variance), and normality of errors for inference. Predictors do not need to be normally
distributed .



Q6: What is the primary purpose of splitting data into training and testing sets?

Answer: To provide an unbiased evaluation of the final model.

Rationale: The testing set is used only to evaluate the final model's performance on unseen
data, providing an unbiased estimate of how the model generalizes to new data .



Q7: In k-fold cross-validation, what is the general purpose of the process?

Answer: To split data into k subsets, using each subset once as a test set and the rest for
training.

Rationale: This provides multiple train-test splits to better estimate model performance and
reduce variance in performance estimates .



Q8: What is overfitting in machine learning?

Answer: When a model performs well on training data but poorly on unseen data.

Rationale: Overfitting occurs when a model learns the training data too well, including noise
and random fluctuations, leading to poor generalization .



Q9: What is the bias-variance tradeoff?

,Answer: Models with high bias tend to underfit, while models with high variance tend
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




to overfit.
k0 k0




Rationale: Simple models have high bias (underfitting) and low variance; complex models have
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




low bias and high variance (overfitting). The goal is to find the right balance .
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




Q10: What is the purpose of regularization in regression models?
k0 k0 k0 k0 k0 k0 k0 k0 k0




k0 Answer: To penalize the size of coefficients and reduce overfitting.
k0 k0 k0 k0 k0 k0 k0 k0 k0




Rationale: Regularization adds a penalty term to the loss function to constrain coefficient
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




magnitudes, helping reduce overfitting and improve generalization .
k0 k0 k0 k0 k0 k0 k0 k0




Q11: What does the Akaike Information Criterion (AIC) balance when comparing models?
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




k0 Answer: Model fit and model complexity.
k0 k0 k0 k0 k0




Rationale: AIC balances model fit (likelihood) with model complexity (number of parameters),
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




penalizing overly complex models to avoid overfitting .
k0 k0 k0 k0 k0 k0 k0 k0




Q12: How does the Bayesian Information Criterion (BIC) differ from AIC?
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




k0 Answer: BIC generally penalizes complexity more heavily than AIC.
k0 k0 k0 k0 k0 k0 k0 k0




Rationale: BIC penalizes model complexity more than AIC, tending to select simpler models,
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




especially with larger sample sizes .
k0 k0 k0 k0 k0 k0




Q13: What is a common disadvantage of k-fold cross-validation?
k0 k0 k0 k0 k0 k0 k0 k0




k0 Answer: It can be computationally expensive for large datasets.
k0 k0 k0 k0 k0 k0 k0 k0




Rationale: K-fold cross-validation requires fitting the model k times, which can be
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




computationally intensive, and it is not directly applicable to time series data without special
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




adaptation .
k0 k0




Q14: LASSO regression can shrink coefficients exactly to zero. True or False?
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




k0 Answer: True. k0

, Rationale: LASSO uses an L1 penalty which can produce sparse solutions, shrinking some
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




coefficients exactly to zero for feature selection .
k0 k0 k0 k0 k0 k0 k0 k0




Q15: Ridge regression uses which type of penalty?
k0 k0 k0 k0 k0 k0 k0




k0 Answer: L2 penalty (sum of squared coefficients).
k0 k0 k0 k0 k0 k0




Rationale: Ridge regression penalizes the sum of squared coefficients, shrinking coefficients
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




toward zero but not exactly to zero, which helps with multicollinearity .
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




Q16: Adjusted R-squared is preferred over R-squared because:
k0 k0 k0 k0 k0 k0 k0




Answer: It penalizes the use of too many variables.
k0 k0 k0 k0 k0 k0 k0 k0




Rationale: Adjusted R-squared is a variant of R² that encourages simpler models by penalizing
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




the addition of unnecessary predictors .
k0 k0 k0 k0 k0 k0




Q17: What is the primary advantage of using a validation set?
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




Answer: It can be used to tune hyperparameters without overfitting to the test set.
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




Rationale: A validation set allows hyperparameter tuning and model selection without touching
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




the final test set, which remains for unbiased evaluation .
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




Q18: What is the consequence of choosing too many variables in a regression model?
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




k0 Answer: The model will likely overfit and have high variance.
k0 k0 k0 k0 k0 k0 k0 k0 k0




Rationale: Including too many predictors increases model complexity, leading to overfitting
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




where the model captures noise rather than true patterns .
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




Q19: What is the curse of dimensionality?
k0 k0 k0 k0 k0 k0




k0 Answer: Data becomes sparse in high dimensions.
k0 k0 k0 k0 k0 k0




Rationale: As the number of dimensions increases, data points become increasingly sparse, and
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0 k0




distance metrics lose meaning, making many algorithms less effective .
k0 k0 k0 k0 k0 k0 k0 k0 k0 k0