data/Stacking.json

{
  "title": "Stacking Mastery: 100 MCQs",
  "description": "A comprehensive set of 100 multiple-choice questions on Stacking ensemble learning, covering basic concepts, implementation, and theoretical understanding.",
  "questions": [
    {
      "id": 1,
      "questionText": "What is the main idea of Stacking in ensemble learning?",
      "options": [
        "Train models in parallel and average results",
        "Train sequential models to reduce bias",
        "Use only one strong learner",
        "Combine predictions of multiple models using a meta-learner"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Stacking involves combining different base learners' predictions with a meta-learner to improve overall performance."
    },
    {
      "id": 2,
      "questionText": "Which component in Stacking combines the outputs of base learners?",
      "options": [
        "Residual estimator",
        "Bootstrap sample",
        "Decision stump",
        "Meta-learner"
      ],
      "correctAnswerIndex": 3,
      "explanation": "The meta-learner takes predictions of base learners as input and produces the final output."
    },
    {
      "id": 3,
      "questionText": "Stacking differs from Bagging because it:",
      "options": [
        "Uses a meta-learner to combine predictions",
        "Only reduces variance",
        "Trains models independently",
        "Uses bootstrapped samples only"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Stacking focuses on learning the best combination of base learners via a meta-model."
    },
    {
      "id": 4,
      "questionText": "Which of the following is a typical base learner in Stacking?",
      "options": [
        "Meta-learner",
        "Feature selector",
        "Residual predictor",
        "Decision tree"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Decision trees, logistic regression, or other models can serve as base learners."
    },
    {
      "id": 5,
      "questionText": "Which of these is a common meta-learner?",
      "options": [
        "Decision stump",
        "Bootstrap sample",
        "Logistic regression",
        "PCA"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Logistic regression or linear regression is often used as a simple meta-learner to combine predictions."
    },
    {
      "id": 6,
      "questionText": "Stacking is most useful when base learners are:",
      "options": [
        "Highly correlated",
        "Identical models",
        "Extremely simple only",
        "Diverse in type or error patterns"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Diversity among base learners allows the meta-learner to exploit complementary strengths."
    },
    {
      "id": 7,
      "questionText": "What is the main advantage of Stacking?",
      "options": [
        "Reduces training time",
        "Improves predictive performance by combining multiple models",
        "Always reduces bias to zero",
        "Eliminates the need for parameter tuning"
      ],
      "correctAnswerIndex": 1,
      "explanation": "By learning from multiple base models, Stacking often achieves higher accuracy than any single model."
    },
    {
      "id": 8,
      "questionText": "In Stacking, which data is used to train the meta-learner?",
      "options": [
        "Original training data only",
        "Residuals of base learners",
        "Randomly generated features",
        "Predictions of base learners on validation or out-of-fold data"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Using out-of-fold predictions prevents overfitting when training the meta-learner."
    },
    {
      "id": 9,
      "questionText": "Which is a difference between Stacking and Boosting?",
      "options": [
        "Stacking reduces variance only",
        "Boosting uses meta-learners, Stacking does not",
        "Stacking combines models in parallel, Boosting sequentially",
        "Boosting uses multiple meta-learners"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Boosting trains models sequentially to correct errors, while Stacking trains models independently and combines their predictions."
    },
    {
      "id": 10,
      "questionText": "Why is cross-validation often used in Stacking?",
      "options": [
        "To select meta-learner automatically",
        "To increase learning rate",
        "To train base learners faster",
        "To generate out-of-fold predictions for training the meta-learner"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Cross-validation provides unbiased predictions of base learners on data not seen during training, which is used to train the meta-learner."
    },
    {
      "id": 11,
      "questionText": "Stacking is also known as:",
      "options": [
        "Random forest ensemble",
        "Boosted regression",
        "Stacked generalization",
        "Sequential bagging"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Stacking was introduced as 'stacked generalization' by Wolpert to combine multiple models."
    },
    {
      "id": 12,
      "questionText": "Which problem does Stacking address that single models might struggle with?",
      "options": [
        "Reducing dataset size",
        "Combining strengths of different algorithms for better generalization",
        "Faster training",
        "Feature scaling"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Stacking leverages different models to capture various patterns and reduce generalization error."
    },
    {
      "id": 13,
      "questionText": "In a classification task, what type of output is passed to the meta-learner?",
      "options": [
        "Random noise",
        "Residuals only",
        "Predicted probabilities or labels from base learners",
        "Original features only"
      ],
      "correctAnswerIndex": 2,
      "explanation": "The meta-learner uses predictions (labels or probabilities) from base learners to make final predictions."
    },
    {
      "id": 14,
      "questionText": "Which is true about the diversity of base learners in Stacking?",
      "options": [
        "All base learners should be identical",
        "Greater diversity usually improves ensemble performance",
        "Meta-learner must be a tree",
        "Only deep trees are used"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Different algorithms or parameter settings increase diversity and help the ensemble learn better."
    },
    {
      "id": 15,
      "questionText": "Which dataset is used to prevent overfitting of the meta-learner?",
      "options": [
        "Random subset of test data",
        "Entire training set predictions",
        "Out-of-fold predictions from training set",
        "Residual errors only"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Out-of-fold predictions give unbiased estimates for the meta-learner to learn safely."
    },
    {
      "id": 16,
      "questionText": "Stacking can be applied to:",
      "options": [
        "Unsupervised tasks only",
        "Both classification and regression tasks",
        "Only classification",
        "Only regression"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Stacking is versatile and can combine base learners for both regression and classification tasks."
    },
    {
      "id": 17,
      "questionText": "Which is NOT a typical base learner in Stacking?",
      "options": [
        "Logistic regression",
        "Decision tree",
        "KNN",
        "Random noise generator"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Random noise is not a meaningful base learner and cannot contribute to ensemble learning."
    },
    {
      "id": 18,
      "questionText": "Meta-learner complexity should be:",
      "options": [
        "Always very deep",
        "Same as base learner complexity",
        "Simple enough to avoid overfitting on base predictions",
        "Randomly selected"
      ],
      "correctAnswerIndex": 2,
      "explanation": "A simple meta-learner generalizes better by learning patterns from base predictions without overfitting."
    },
    {
      "id": 19,
      "questionText": "Which of the following can be used as meta-learner?",
      "options": [
        "Random features only",
        "Bootstrap samples",
        "Noise vector",
        "Linear regression, logistic regression, or tree"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Various models can serve as meta-learner depending on the problem type."
    },
    {
      "id": 20,
      "questionText": "Stacking usually improves performance when base learners:",
      "options": [
        "Have complementary strengths and weaknesses",
        "Are identical in type",
        "Have zero diversity",
        "Are only weak learners"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Combining models with different strengths allows the meta-learner to correct errors and improve predictions."
    },
    {
      "id": 21,
      "questionText": "Which is a common mistake when implementing Stacking?",
      "options": [
        "Using simple meta-learner",
        "Using cross-validation for base predictions",
        "Training meta-learner on same data base learners saw",
        "Using different base learners"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Training meta-learner on same data can cause overfitting; out-of-fold predictions prevent this."
    },
    {
      "id": 22,
      "questionText": "Stacking differs from Voting because:",
      "options": [
        "It reduces variance only",
        "It averages predictions blindly",
        "It learns weights using a meta-learner rather than using fixed rules",
        "It uses bootstrap samples only"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Unlike Voting, Stacking trains a model to optimally combine base learners’ predictions."
    },
    {
      "id": 23,
      "questionText": "Which scenario benefits most from Stacking?",
      "options": [
        "Identical models only",
        "Single model with high accuracy",
        "When multiple different models have complementary predictive power",
        "Very small datasets"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Stacking leverages diverse models to produce better generalization than any individual model."
    },
    {
      "id": 24,
      "questionText": "Which metric should you use to evaluate Stacking?",
      "options": [
        "Depends on the problem (accuracy, RMSE, F1, etc.)",
        "Always F1-score",
        "Always RMSE",
        "Always accuracy"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Evaluation metric depends on the type of task (classification or regression)."
    },
    {
      "id": 25,
      "questionText": "In K-fold Stacking, each fold provides predictions to:",
      "options": [
        "Train the meta-learner without overfitting",
        "Generate residuals",
        "Train base learners only",
        "Randomly select features"
      ],
      "correctAnswerIndex": 0,
      "explanation": "K-fold cross-validation provides unbiased predictions from base learners for the meta-learner."
    },
    {
      "id": 26,
      "questionText": "Stacking can reduce generalization error by:",
      "options": [
        "Randomly averaging predictions",
        "Ignoring base learners",
        "Combining strengths of multiple models",
        "Using only a single strong model"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Meta-learner exploits complementary strengths of base learners to improve predictions."
    },
    {
      "id": 27,
      "questionText": "Which is true for regression tasks using Stacking?",
      "options": [
        "Meta-learner predicts labels only",
        "Only classification is possible",
        "Residuals are ignored",
        "Base learners predict continuous values, meta-learner combines them"
      ],
      "correctAnswerIndex": 3,
      "explanation": "For regression, the meta-learner learns to combine continuous predictions from base learners."
    },
    {
      "id": 28,
      "questionText": "Which prevents overfitting in Stacking?",
      "options": [
        "Ignoring diversity of base learners",
        "Deep meta-learner only",
        "Using out-of-fold predictions for meta-learner training",
        "Training meta-learner on entire dataset predictions"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Out-of-fold predictions prevent the meta-learner from memorizing base learners’ predictions."
    },
    {
      "id": 29,
      "questionText": "Scenario: Combining Random Forest, SVM, and KNN with a linear meta-learner. This is:",
      "options": [
        "Boosting",
        "Bagging",
        "Stacking",
        "Voting"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Different base learners are combined via a meta-learner, which defines Stacking."
    },
    {
      "id": 30,
      "questionText": "Which is the main requirement for base learners in Stacking?",
      "options": [
        "They must be deep trees only",
        "They should be diverse and not perfectly correlated",
        "They should always be linear models",
        "They must have identical predictions"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Diversity ensures that the meta-learner can learn from complementary strengths of different models."
    },
    {
      "id": 31,
      "questionText": "In Stacking, why is it important that base learners are diverse?",
      "options": [
        "Identical base learners are always better",
        "Diversity increases bias",
        "Diverse base learners capture different aspects of the data, improving meta-learner performance",
        "Diversity reduces computation"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Diversity among base learners ensures complementary strengths, which the meta-learner can exploit for better predictions."
    },
    {
      "id": 32,
      "questionText": "Which technique is commonly used to generate unbiased predictions for meta-learner training?",
      "options": [
        "K-fold cross-validation (out-of-fold predictions)",
        "Random feature selection",
        "Using test data",
        "Bootstrap sampling only"
      ],
      "correctAnswerIndex": 0,
      "explanation": "K-fold cross-validation produces predictions from unseen data folds to prevent overfitting when training the meta-learner."
    },
    {
      "id": 33,
      "questionText": "Scenario: You use three base learners with high correlation. What is likely to happen?",
      "options": [
        "The meta-learner ignores correlation automatically",
        "Performance will drastically improve",
        "Overfitting is impossible",
        "The meta-learner gains little benefit due to redundant information"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Highly correlated base learners do not provide complementary information, reducing the benefit of Stacking."
    },
    {
      "id": 34,
      "questionText": "Which type of meta-learner is commonly used for regression tasks?",
      "options": [
        "Decision stump",
        "Logistic regression",
        "Linear regression or ridge regression",
        "Random noise generator"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Linear or regularized regression models are simple and effective for combining continuous outputs of base learners."
    },
    {
      "id": 35,
      "questionText": "Which type of meta-learner is commonly used for classification tasks?",
      "options": [
        "K-means clustering",
        "Random noise generator",
        "Logistic regression",
        "Linear regression"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Logistic regression can combine probability outputs from base learners and produce final class probabilities."
    },
    {
      "id": 36,
      "questionText": "Stacking can be applied to:",
      "options": [
        "Classification and regression",
        "Unsupervised tasks only",
        "Only classification",
        "Only regression"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Stacking is versatile and works for both classification and regression problems."
    },
    {
      "id": 37,
      "questionText": "Scenario: Base learners perform poorly individually but differently. Stacking may:",
      "options": [
        "Always fail",
        "Reduce bias only",
        "Increase correlation among predictions",
        "Improve overall performance by combining diverse predictions"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Even weak base learners can be combined effectively by the meta-learner if they make different errors."
    },
    {
      "id": 38,
      "questionText": "Why should meta-learner complexity be limited?",
      "options": [
        "To prevent overfitting on base learners’ predictions",
        "To reduce dataset size",
        "To increase training time",
        "Because base learners are always simple"
      ],
      "correctAnswerIndex": 0,
      "explanation": "A simple meta-learner generalizes better on predictions from base learners without memorizing noise."
    },
    {
      "id": 39,
      "questionText": "Scenario: Using Random Forest, SVM, and KNN as base learners with Logistic Regression as meta-learner. Which is true?",
      "options": [
        "Diverse base learners + simple meta-learner is a common Stacking setup",
        "Base learners must be identical",
        "Meta-learner should be very deep",
        "Only regression problems are supported"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Combining different algorithms with a simple meta-learner is a standard approach in Stacking."
    },
    {
      "id": 40,
      "questionText": "Scenario: Your meta-learner overfits the base learners’ predictions. Which solution is suitable?",
      "options": [
        "Use simpler meta-learner or regularization",
        "Add more base learners without change",
        "Increase base learner complexity",
        "Ignore cross-validation"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Regularizing or simplifying the meta-learner reduces overfitting on base predictions."
    },
    {
      "id": 41,
      "questionText": "Which cross-validation strategy is used to generate predictions for meta-learner training?",
      "options": [
        "Random sampling",
        "No CV is needed",
        "K-fold cross-validation",
        "Leave-one-out only"
      ],
      "correctAnswerIndex": 2,
      "explanation": "K-fold CV produces out-of-fold predictions to prevent overfitting of the meta-learner."
    },
    {
      "id": 42,
      "questionText": "Stacking differs from Voting because:",
      "options": [
        "It learns combination weights via a meta-learner",
        "It reduces variance only",
        "It uses identical base learners",
        "It averages predictions blindly"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Voting combines base learners using fixed rules, while Stacking learns how to combine predictions optimally."
    },
    {
      "id": 43,
      "questionText": "Scenario: Your dataset is small. Stacking may:",
      "options": [
        "Always improve accuracy",
        "Overfit due to limited training data for meta-learner",
        "Reduce computation time automatically",
        "Ignore base learners"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Meta-learner may overfit if there isn’t enough data for unbiased predictions from base learners."
    },
    {
      "id": 44,
      "questionText": "Which situation is ideal for using Stacking?",
      "options": [
        "Highly correlated base learners",
        "No training data available",
        "Single strong model is sufficient",
        "Multiple different models have complementary strengths"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Stacking benefits when base learners make different types of errors, allowing meta-learner to combine them effectively."
    },
    {
      "id": 45,
      "questionText": "Why are out-of-fold predictions used instead of training predictions for the meta-learner?",
      "options": [
        "To add noise intentionally",
        "To prevent meta-learner from overfitting",
        "To reduce computation",
        "To increase correlation"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Using predictions on unseen folds ensures the meta-learner sees unbiased predictions and generalizes better."
    },
    {
      "id": 46,
      "questionText": "Scenario: All base learners are trees with same depth. How to improve stacking?",
      "options": [
        "Use only meta-learner",
        "Add more identical trees",
        "Reduce training data",
        "Increase diversity via different algorithms or hyperparameters"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Diverse learners are key for stacking; otherwise, meta-learner gains little new information."
    },
    {
      "id": 47,
      "questionText": "Which of the following helps prevent overfitting in stacking?",
      "options": [
        "Adding noise to predictions",
        "Deep meta-learner only",
        "High learning rate only",
        "Cross-validation, simpler meta-learner, regularization"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Using CV and regularization ensures meta-learner does not memorize base learners’ predictions."
    },
    {
      "id": 48,
      "questionText": "Which task is stacking suitable for?",
      "options": [
        "Structured regression, classification, and hybrid tasks",
        "Only unsupervised learning",
        "Only image generation",
        "Only dimensionality reduction"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Stacking is versatile and can be applied to any supervised task."
    },
    {
      "id": 49,
      "questionText": "Scenario: You want to combine a Random Forest and a KNN for classification. What is a suitable meta-learner?",
      "options": [
        "Logistic regression",
        "K-means clustering",
        "Principal Component Analysis",
        "Another Random Forest only"
      ],
      "correctAnswerIndex": 0,
      "explanation": "A simple model like logistic regression can effectively combine predictions from heterogeneous base learners."
    },
    {
      "id": 50,
      "questionText": "Why is meta-learner training data usually smaller than base learner training data?",
      "options": [
        "It sees random features only",
        "It uses the entire dataset again",
        "It only sees residuals",
        "It uses out-of-fold predictions from base learners"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Meta-learner sees predictions on validation folds, not full training data, to avoid overfitting."
    },
    {
      "id": 51,
      "questionText": "Scenario: Base learners predict different class probabilities for a sample. What does the meta-learner do?",
      "options": [
        "Selects the first base learner only",
        "Combines these predictions to make the final decision",
        "Averages features instead of predictions",
        "Ignores all predictions"
      ],
      "correctAnswerIndex": 1,
      "explanation": "The meta-learner uses outputs from base learners as inputs to produce a more accurate final prediction."
    },
    {
      "id": 52,
      "questionText": "Which of these is a benefit of using Stacking over individual models?",
      "options": [
        "Reduces dataset size automatically",
        "Improved predictive performance by combining strengths of multiple models",
        "Always faster training",
        "No need for cross-validation"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Stacking leverages diverse models to capture different patterns and reduce overall error."
    },
    {
      "id": 53,
      "questionText": "Scenario: Stacking with highly correlated base learners results in:",
      "options": [
        "Limited improvement due to redundant predictions",
        "No need for a meta-learner",
        "Automatic error correction",
        "Maximum improvement always"
      ],
      "correctAnswerIndex": 0,
      "explanation": "If base learners make similar errors, the meta-learner gains little new information."
    },
    {
      "id": 54,
      "questionText": "Which factor is crucial for effective Stacking?",
      "options": [
        "Training base learners on same features only",
        "Identical predictions from all base learners",
        "Diversity among base learners",
        "Using a deep meta-learner only"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Different algorithms or parameters ensure base learners capture complementary information."
    },
    {
      "id": 55,
      "questionText": "Scenario: Small dataset, multiple base learners. Meta-learner shows overfitting. Recommended solution?",
      "options": [
        "Increase number of trees only",
        "Ignore cross-validation",
        "Increase meta-learner complexity",
        "Use simpler meta-learner or regularization, possibly reduce number of base learners"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Simpler meta-learner and regularization prevent overfitting when training data is limited."
    },
    {
      "id": 56,
      "questionText": "Why is stacking preferred over simple averaging or voting in some cases?",
      "options": [
        "It always uses deep learning",
        "It learns optimal weights for combining predictions instead of using fixed rules",
        "It eliminates need for base learners",
        "It reduces computation time"
      ],
      "correctAnswerIndex": 1,
      "explanation": "The meta-learner can adaptively combine base predictions based on data patterns, improving accuracy."
    },
    {
      "id": 57,
      "questionText": "Scenario: Base learners are decision trees with shallow depth. Meta-learner is logistic regression. Likely effect?",
      "options": [
        "Meta-learner can capture complementary signals and improve performance",
        "Performance will always drop",
        "Trees become irrelevant",
        "Only overfitting occurs"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Even weak or shallow learners can provide useful signals for the meta-learner."
    },
    {
      "id": 58,
      "questionText": "Which is a common mistake in Stacking implementation?",
      "options": [
        "Using simple meta-learner",
        "Training meta-learner on base learners’ training predictions (not out-of-fold predictions)",
        "Using diverse base learners",
        "Cross-validation for base predictions"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Using training predictions directly can cause overfitting; out-of-fold predictions are needed."
    },
    {
      "id": 59,
      "questionText": "Scenario: Stacking regression with three base learners. Which output type does the meta-learner use?",
      "options": [
        "Predicted classes only",
        "Random noise vector",
        "Residuals only",
        "Predicted continuous values from base learners"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Meta-learner combines predicted continuous outputs from base learners to produce final regression output."
    },
    {
      "id": 60,
      "questionText": "Scenario: You have Random Forest, XGBoost, and SVM as base learners. Which meta-learner is simple and effective?",
      "options": [
        "PCA",
        "Deep neural network only",
        "Logistic regression or linear regression",
        "Random noise generator"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Simple regression models can effectively combine heterogeneous predictions without overfitting."
    },
    {
      "id": 61,
      "questionText": "Scenario: Meta-learner predicts perfectly on training data but poorly on test data. Cause?",
      "options": [
        "Dataset too large",
        "Meta-learner too simple",
        "Overfitting due to using training predictions instead of out-of-fold predictions",
        "Base learners are too diverse"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Training on base learners’ predictions from the same data leads to memorization and poor generalization."
    },
    {
      "id": 62,
      "questionText": "Which of these is NOT a recommended strategy in Stacking?",
      "options": [
        "Using out-of-fold predictions",
        "Using cross-validation for base learners",
        "Regularizing the meta-learner",
        "Using meta-learner trained on base learners’ training predictions"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Meta-learner must be trained on unbiased predictions; using training predictions causes overfitting."
    },
    {
      "id": 63,
      "questionText": "Scenario: Base learners have high variance individually. Stacking can:",
      "options": [
        "Always increase bias",
        "Reduce overall variance by combining their predictions",
        "Ignore base learner predictions",
        "Eliminate need for cross-validation"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Meta-learner can combine different noisy predictions to reduce overall variance and improve stability."
    },
    {
      "id": 64,
      "questionText": "Scenario: Base learners are homogeneous (e.g., all logistic regressions). Likely effect?",
      "options": [
        "Meta-learner ignored",
        "Maximum benefit always",
        "Overfitting impossible",
        "Limited improvement from Stacking due to redundancy"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Stacking works best when base learners are diverse; homogeneous models provide little new information."
    },
    {
      "id": 65,
      "questionText": "Which approach improves stacking with limited data?",
      "options": [
        "More complex meta-learner only",
        "Ignore base learner diversity",
        "Regularization, simpler meta-learner, careful cross-validation",
        "Train meta-learner on training predictions"
      ],
      "correctAnswerIndex": 2,
      "explanation": "These strategies reduce overfitting and improve generalization when data is scarce."
    },
    {
      "id": 66,
      "questionText": "Scenario: Meta-learner underfits base predictions. Recommended fix?",
      "options": [
        "Use training predictions instead of out-of-fold",
        "Reduce base learner diversity",
        "Use a slightly more complex meta-learner or additional features",
        "Ignore predictions"
      ],
      "correctAnswerIndex": 2,
      "explanation": "A slightly more flexible meta-learner can better capture relationships between base learners’ predictions."
    },
    {
      "id": 67,
      "questionText": "Scenario: Combining Random Forest and Gradient Boosting as base learners. Which advantage does stacking provide?",
      "options": [
        "Eliminates bias automatically",
        "Leverages complementary strengths of ensemble methods for better prediction",
        "Reduces variance to zero",
        "Replaces base learners completely"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Stacking allows different ensembles to complement each other, improving overall performance."
    },
    {
      "id": 68,
      "questionText": "Scenario: Using stacking in classification, base learners predict probabilities. Meta-learner input?",
      "options": [
        "Random noise vector",
        "Predicted probabilities from base learners",
        "Original features only",
        "Residual errors only"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Meta-learner uses predicted probabilities from base learners as inputs to produce final classification."
    },
    {
      "id": 69,
      "questionText": "Which scenario would reduce the benefit of stacking?",
      "options": [
        "Base learners are diverse",
        "Base learners are highly correlated",
        "Out-of-fold predictions are used",
        "Meta-learner is regularized"
      ],
      "correctAnswerIndex": 1,
      "explanation": "High correlation among base learners provides redundant information, limiting stacking’s advantage."
    },
    {
      "id": 70,
      "questionText": "Scenario: Stacking regression task shows overfitting. First check:",
      "options": [
        "Whether meta-learner was trained on out-of-fold predictions",
        "Base learner type only",
        "Number of features only",
        "Dataset size only"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Using training predictions instead of out-of-fold predictions is a common cause of overfitting in stacking."
    },
    {
      "id": 71,
      "questionText": "Scenario: In a Kaggle competition, you combine multiple tree-based and linear models. Your meta-learner performs worse than individual base learners. Likely cause?",
      "options": [
        "Base learners are too diverse",
        "Dataset is too large",
        "Meta-learner overfitted due to training on base learners’ training predictions",
        "Meta-learner is too simple"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Training the meta-learner on the same data as base learners can cause memorization and poor generalization."
    },
    {
      "id": 72,
      "questionText": "Scenario: You notice highly correlated predictions from base learners. Which action is appropriate?",
      "options": [
        "Ignore the correlation",
        "Increase number of trees in all learners",
        "Introduce more diverse base learners",
        "Use the same algorithm with different hyperparameters only"
      ],
      "correctAnswerIndex": 2,
      "explanation": "High correlation reduces the benefit of stacking. Introducing diverse models captures complementary patterns."
    },
    {
      "id": 73,
      "questionText": "Scenario: Base learners are neural networks with slightly different architectures. Meta-learner is linear regression. What is expected?",
      "options": [
        "Meta-learner can combine complementary predictions to improve accuracy",
        "Performance always decreases",
        "Meta-learner will ignore base learners",
        "Stacking will fail because linear models cannot handle neural networks"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Linear meta-learner can learn optimal weights for combining diverse neural network outputs."
    },
    {
      "id": 74,
      "questionText": "Scenario: Using stacking for regression, meta-learner outputs extreme values. Cause?",
      "options": [
        "Base learners’ predictions are poorly scaled or meta-learner is too complex",
        "Base learners are too diverse",
        "Meta-learner underfitted",
        "Dataset is too small"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Improper scaling or an overly complex meta-learner can lead to extreme predictions."
    },
    {
      "id": 75,
      "questionText": "Scenario: You stack three models and notice high variance in meta-learner. Solution?",
      "options": [
        "Add more identical base learners",
        "Regularize meta-learner or reduce complexity",
        "Ignore variance",
        "Use training predictions instead of out-of-fold"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Regularization prevents meta-learner from overfitting to noisy base learner predictions."
    },
    {
      "id": 76,
      "questionText": "Scenario: Base learners perform poorly individually but differently. Stacking improves results. Why?",
      "options": [
        "Base learners are ignored",
        "Meta-learner leverages complementary errors to produce better overall predictions",
        "Stacking magically improves all models",
        "Random averaging occurs"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Even weak but diverse models can be combined effectively by the meta-learner."
    },
    {
      "id": 77,
      "questionText": "Scenario: Meta-learner is too powerful (e.g., deep neural network). What is the likely outcome?",
      "options": [
        "Improved generalization automatically",
        "Overfitting to base learners’ predictions",
        "Dataset size decreases",
        "Base learners become irrelevant"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Overly complex meta-learner may memorize base predictions instead of learning patterns, leading to poor generalization."
    },
    {
      "id": 78,
      "questionText": "Scenario: Small dataset with many base learners. Meta-learner underfits. Solution?",
      "options": [
        "Reduce base learner complexity or number",
        "Train on test data",
        "Ignore diversity",
        "Increase meta-learner complexity"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Too many base learners can overwhelm meta-learner on small datasets. Reducing base learners or their complexity helps."
    },
    {
      "id": 79,
      "questionText": "Scenario: Regression stacking task shows systematic bias. Solution?",
      "options": [
        "Adjust meta-learner to correct bias or apply residual correction",
        "Use training predictions instead of out-of-fold",
        "Increase number of base learners only",
        "Ignore base learners"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Meta-learner can be tuned or trained on residuals to correct systematic bias."
    },
    {
      "id": 80,
      "questionText": "Scenario: Ensemble includes Random Forest, XGBoost, and KNN. Test accuracy decreases after stacking. First check?",
      "options": [
        "Whether meta-learner was trained on proper out-of-fold predictions",
        "Number of trees only",
        "Feature selection only",
        "Dataset size only"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Improper meta-learner training is the most common cause of poor stacking performance."
    },
    {
      "id": 81,
      "questionText": "Scenario: You want to combine multiple image classifiers via stacking. Which approach is suitable?",
      "options": [
        "Use softmax probabilities from base classifiers as meta-learner input",
        "Use raw pixel inputs",
        "Ignore base classifiers",
        "Average features randomly"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Meta-learner combines probability predictions rather than raw data for effective stacking."
    },
    {
      "id": 82,
      "questionText": "Scenario: In a stacking setup, meta-learner shows perfect training accuracy. Likely issue?",
      "options": [
        "Overfitting due to using base learners’ training predictions",
        "Base learners are too diverse",
        "Meta-learner too simple",
        "Dataset too small"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Perfect training accuracy is a sign of overfitting; out-of-fold predictions prevent this."
    },
    {
      "id": 83,
      "questionText": "Scenario: Base learners are all SVMs with different kernels. Meta-learner is logistic regression. Likely outcome?",
      "options": [
        "Improved generalization due to diversity in kernel functions",
        "No improvement, identical predictions",
        "Overfitting impossible",
        "Meta-learner ignored"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Different kernels capture complementary patterns, allowing meta-learner to improve predictions."
    },
    {
      "id": 84,
      "questionText": "Scenario: Base learners have high variance errors. Stacking improves predictions. Why?",
      "options": [
        "Meta-learner combines predictions to reduce variance and improve stability",
        "Stacking magically reduces errors",
        "Base learners are ignored",
        "Random averaging occurs"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Meta-learner can smooth out high variance by learning the optimal combination of predictions."
    },
    {
      "id": 85,
      "questionText": "Scenario: Regression stacking task shows systematic bias. Solution?",
      "options": [
        "Adjust meta-learner to correct bias or apply residual correction",
        "Ignore base learners",
        "Increase number of base learners only",
        "Use training predictions instead of out-of-fold"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Meta-learner can be tuned or trained on residuals to correct systematic bias."
    },
    {
      "id": 86,
      "questionText": "Scenario: Base learners predict probabilities for multi-class classification. Meta-learner input?",
      "options": [
        "Concatenated class probabilities from all base learners",
        "Raw features only",
        "Residuals only",
        "Random noise vector"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Meta-learner uses predicted probabilities from all classes to make the final decision."
    },
    {
      "id": 87,
      "questionText": "Scenario: Meta-learner underfits in a classification stacking task. Recommended action?",
      "options": [
        "Increase meta-learner capacity slightly or add engineered features",
        "Reduce base learner diversity",
        "Ignore base learners",
        "Train meta-learner on training predictions"
      ],
      "correctAnswerIndex": 0,
      "explanation": "A slightly more complex meta-learner can capture relationships between base learners’ outputs."
    },
    {
      "id": 88,
      "questionText": "Scenario: Small dataset, multiple base learners. Meta-learner overfits. Best solution?",
      "options": [
        "Use simpler meta-learner and regularization",
        "Add more base learners",
        "Ignore cross-validation",
        "Train meta-learner on training predictions"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Simpler meta-learner with regularization prevents overfitting on limited out-of-fold predictions."
    },
    {
      "id": 89,
      "questionText": "Scenario: Base learners include gradient boosting, random forest, and logistic regression. Stacking improves performance. Why?",
      "options": [
        "Meta-learner exploits complementary predictions of heterogeneous models",
        "Stacking magically improves results",
        "Base learners are ignored",
        "Dataset size increases"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Diverse models capture different patterns, which meta-learner combines for better generalization."
    },
    {
      "id": 90,
      "questionText": "Scenario: You want to stack deep learning models for regression. Best approach?",
      "options": [
        "Use predicted outputs or features from penultimate layers as meta-learner input",
        "Raw images only",
        "Ignore base learners",
        "Average base model weights"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Using predictions or embeddings from deep models is standard for stacking to combine outputs effectively."
    },
    {
      "id": 91,
      "questionText": "Scenario: Base learners are overfitting slightly. Meta-learner underfits. Recommendation?",
      "options": [
        "Reduce base learner overfitting and slightly increase meta-learner capacity",
        "Ignore base learners",
        "Train meta-learner on test data",
        "Increase dataset size only"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Balancing base and meta-learner capacities improves overall stacking performance."
    },
    {
      "id": 92,
      "questionText": "Scenario: Stacking regression, meta-learner predicts negative values where base predictions are positive. Fix?",
      "options": [
        "Check scaling and bias adjustments in meta-learner",
        "Ignore predictions",
        "Reduce base learners",
        "Use training predictions instead of out-of-fold"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Meta-learner may require proper scaling or offset to combine base predictions correctly."
    },
    {
      "id": 93,
      "questionText": "Scenario: Meta-learner training time is extremely high. Possible solution?",
      "options": [
        "Reduce number of base learners or use simpler meta-learner",
        "Increase base learner complexity",
        "Ignore training time",
        "Use training predictions directly"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Simplifying the meta-learner or reducing base learners can significantly lower computation time."
    },
    {
      "id": 94,
      "questionText": "Scenario: Stacking for imbalanced classification. Recommended approach?",
      "options": [
        "Use probability outputs and apply class weighting or sampling strategies",
        "Ignore imbalance",
        "Train meta-learner on majority class only",
        "Use raw features directly"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Meta-learner can be trained with balanced inputs to handle imbalanced datasets effectively."
    },
    {
      "id": 95,
      "questionText": "Scenario: Multiple base learners provide continuous outputs with different scales. What is recommended?",
      "options": [
        "Normalize or standardize outputs before feeding into meta-learner",
        "Ignore scale differences",
        "Train meta-learner on raw values",
        "Use only one base learner"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Meta-learner performs better when inputs are on comparable scales."
    },
    {
      "id": 96,
      "questionText": "Scenario: Stacking with three classifiers, meta-learner predicts incorrectly on edge cases. Solution?",
      "options": [
        "Use more diverse base learners or add engineered features",
        "Reduce base learner diversity",
        "Ignore predictions",
        "Train on training predictions only"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Meta-learner can improve predictions on edge cases if base learners provide complementary information."
    },
    {
      "id": 97,
      "questionText": "Scenario: You stack tree-based models with logistic regression meta-learner. Test RMSE is higher than best base learner. Likely cause?",
      "options": [
        "Meta-learner overfitted or base predictions too correlated",
        "Stacking always reduces RMSE",
        "Dataset too large",
        "Meta-learner too simple"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Correlation among base learners or overfitting in meta-learner can degrade performance."
    },
    {
      "id": 98,
      "questionText": "Scenario: Combining heterogeneous models via stacking for regression. Key considerations?",
      "options": [
        "Diversity, proper meta-learner training, scaling of outputs",
        "Use identical base learners only",
        "Ignore cross-validation",
        "Increase number of base learners blindly"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Effective stacking requires diverse base learners, out-of-fold meta-learner training, and proper scaling."
    },
    {
      "id": 99,
      "questionText": "Scenario: Meta-learner underfits in a classification stacking task. Recommended action?",
      "options": [
        "Increase meta-learner capacity slightly or add engineered features",
        "Reduce base learner diversity",
        "Ignore base learners",
        "Train meta-learner on training predictions"
      ],
      "correctAnswerIndex": 0,
      "explanation": "A slightly more complex meta-learner can capture relationships between base learners’ outputs."
    },
    {
      "id": 100,
      "questionText": "Scenario: Stacking regression ensemble shows overfitting. Which step should be prioritized?",
      "options": [
        "Verify meta-learner uses out-of-fold predictions and apply regularization",
        "Add more base learners",
        "Ignore overfitting",
        "Train meta-learner on full training predictions"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Out-of-fold predictions and regularization are essential to prevent overfitting in stacking ensembles."
    }
  ]
}