{
  "title": "Gradient Boosting Mastery: 100 MCQs",
  "description": "A complete 100-question set to master Gradient Boosting — covering boosting basics, weak learners, sequential correction, advanced hyperparameters, regularization, and real-world scenarios.",
  "questions": [
    {
      "id": 1,
      "questionText": "What is the main idea behind Gradient Boosting?",
      "options": [
        "Reduce dimensions before training",
        "Use only one deep decision tree",
        "Sequentially build models to correct errors of previous ones",
        "Combine models randomly"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Gradient Boosting builds models sequentially, with each new model correcting the errors made by the previous ensemble."
    },
    {
      "id": 2,
      "questionText": "Which type of learner is typically used in Gradient Boosting?",
      "options": [
        "PCA components",
        "K-Means clusters",
        "Neural Networks",
        "Decision Trees"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Gradient Boosting commonly uses shallow decision trees (weak learners) for sequential correction."
    },
    {
      "id": 3,
      "questionText": "Scenario: You notice your Gradient Boosting model is underfitting. Which action could help?",
      "options": [
        "Use fewer estimators",
        "Reduce dataset size",
        "Reduce learning rate",
        "Increase tree depth"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Increasing tree depth allows each weak learner to capture more complex patterns, reducing underfitting."
    },
    {
      "id": 4,
      "questionText": "What role does the learning rate play in Gradient Boosting?",
      "options": [
        "Controls tree pruning only",
        "Controls the contribution of each tree to the ensemble",
        "Controls the number of features",
        "Controls the dataset size"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Learning rate scales the contribution of each tree; lower values slow learning and improve generalization."
    },
    {
      "id": 5,
      "questionText": "Scenario: Your Gradient Boosting model has perfect training accuracy but poor test accuracy. What is likely happening?",
      "options": [
        "High bias",
        "Optimal fit",
        "Overfitting",
        "Underfitting"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Perfect training accuracy with poor generalization indicates overfitting."
    },
    {
      "id": 6,
      "questionText": "Which metric is commonly minimized by Gradient Boosting?",
      "options": [
        "Confusion matrix values",
        "Accuracy",
        "Loss (cost) function",
        "F1-score"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Gradient Boosting minimizes a differentiable loss function using gradient descent in function space."
    },
    {
      "id": 7,
      "questionText": "Scenario: You want to speed up Gradient Boosting without losing much accuracy. Which technique helps?",
      "options": [
        "Reduce number of trees to 1",
        "Use very deep trees",
        "Increase learning rate drastically",
        "Subsampling (stochastic gradient boosting)"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Subsampling (training on a random subset per iteration) speeds up computation and can improve generalization."
    },
    {
      "id": 8,
      "questionText": "What is the primary benefit of using weak learners in Gradient Boosting?",
      "options": [
        "They achieve perfect predictions alone",
        "They reduce computation and allow sequential correction",
        "They perform clustering",
        "They reduce data dimensionality"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Weak learners are simple models that improve performance when combined sequentially."
    },
    {
      "id": 9,
      "questionText": "Scenario: Your Gradient Boosting model is sensitive to outliers. What is a common solution?",
      "options": [
        "Reduce learning rate to zero",
        "Increase tree depth",
        "Use robust loss functions like Huber loss",
        "Use only one estimator"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Robust loss functions reduce the influence of outliers on the model."
    },
    {
      "id": 10,
      "questionText": "Which parameter controls the maximum number of trees in Gradient Boosting?",
      "options": [
        "max_depth",
        "subsample",
        "learning_rate",
        "n_estimators"
      ],
      "correctAnswerIndex": 3,
      "explanation": "The n_estimators parameter specifies how many sequential trees are built in the ensemble."
    },
    {
      "id": 11,
      "questionText": "Scenario: You increase the number of estimators but leave learning rate high. What is likely to happen?",
      "options": [
        "Underfitting decreases",
        "Nothing significant",
        "Overfitting may increase",
        "Model accuracy drops immediately"
      ],
      "correctAnswerIndex": 2,
      "explanation": "A high learning rate with many trees can cause overfitting since each tree contributes too much."
    },
    {
      "id": 12,
      "questionText": "What is the role of the residual in Gradient Boosting?",
      "options": [
        "Represents feature importance",
        "Represents total variance",
        "Represents errors from previous models to be corrected",
        "Represents learning rate"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Residuals are the differences between predicted and actual values; new trees aim to predict them."
    },
    {
      "id": 13,
      "questionText": "Scenario: Training Gradient Boosting with very deep trees. Risk?",
      "options": [
        "High bias",
        "Reduced training time",
        "Underfitting",
        "Overfitting"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Very deep trees can overfit the training data and generalize poorly."
    },
    {
      "id": 14,
      "questionText": "Which of the following is a key advantage of Gradient Boosting over single Decision Trees?",
      "options": [
        "Higher predictive accuracy",
        "Handles missing data automatically",
        "No hyperparameters",
        "Less computation"
      ],
      "correctAnswerIndex": 0,
      "explanation": "By combining many weak learners sequentially, Gradient Boosting improves prediction accuracy compared to a single tree."
    },
    {
      "id": 15,
      "questionText": "Scenario: You are using Gradient Boosting with imbalanced classes. Recommended approach?",
      "options": [
        "Ignore the imbalance",
        "Increase learning rate",
        "Use class weights or specialized loss functions",
        "Reduce number of trees"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Class weighting or modified loss functions helps Gradient Boosting handle class imbalance effectively."
    },
    {
      "id": 16,
      "questionText": "Which of the following is NOT a common loss function for Gradient Boosting?",
      "options": [
        "Deviance (logistic loss)",
        "Least squares regression",
        "Euclidean distance",
        "Huber loss"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Euclidean distance is not used directly as a loss function; least squares, deviance, and Huber loss are standard choices."
    },
    {
      "id": 17,
      "questionText": "Scenario: You reduce learning rate too much while keeping n_estimators small. Effect?",
      "options": [
        "Overfitting",
        "Random predictions",
        "Immediate convergence",
        "Underfitting due to slow learning"
      ],
      "correctAnswerIndex": 3,
      "explanation": "A very low learning rate with few trees may prevent the model from fitting the data sufficiently, causing underfitting."
    },
    {
      "id": 18,
      "questionText": "What is the difference between Gradient Boosting and AdaBoost?",
      "options": [
        "They are identical",
        "Gradient Boosting optimizes a differentiable loss; AdaBoost adjusts weights on misclassified samples",
        "AdaBoost uses neural networks; Gradient Boosting uses trees",
        "Gradient Boosting is unsupervised"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Gradient Boosting minimizes a loss function via gradients; AdaBoost focuses on weighting misclassified examples."
    },
    {
      "id": 19,
      "questionText": "Scenario: You have noisy data. Which adjustment helps Gradient Boosting perform better?",
      "options": [
        "Increase tree depth aggressively",
        "Lower learning rate and smaller tree depth",
        "Increase learning rate",
        "Reduce number of features"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Lower learning rate and shallower trees prevent the model from fitting noise in the data."
    },
    {
      "id": 20,
      "questionText": "Which parameter controls the randomness of rows sampled per tree?",
      "options": [
        "learning_rate",
        "n_estimators",
        "max_depth",
        "subsample"
      ],
      "correctAnswerIndex": 3,
      "explanation": "The subsample parameter specifies the fraction of rows used per iteration, introducing randomness and helping generalization."
    },
    {
      "id": 21,
      "questionText": "Scenario: Gradient Boosting is slow on a large dataset. Possible solution besides subsampling?",
      "options": [
        "Add more trees",
        "Use deeper trees",
        "Increase learning rate drastically",
        "Reduce max_depth or min_samples_split"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Shallower trees and stricter splitting criteria reduce computation per tree and speed up training."
    },
    {
      "id": 22,
      "questionText": "What is the effect of increasing the number of estimators while keeping learning rate constant?",
      "options": [
        "Reduced training time",
        "Learning rate becomes irrelevant",
        "Model may overfit if learning rate is high",
        "Underfitting"
      ],
      "correctAnswerIndex": 2,
      "explanation": "More estimators increase model capacity; with high learning rate, overfitting is more likely."
    },
    {
      "id": 23,
      "questionText": "Scenario: You want to use Gradient Boosting for regression. Which loss function is typical?",
      "options": [
        "Least squares (MSE)",
        "Log loss",
        "Cross-entropy",
        "Hinge loss"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Mean squared error (least squares) is standard for regression tasks in Gradient Boosting."
    },
    {
      "id": 24,
      "questionText": "Which technique helps Gradient Boosting handle high-dimensional datasets?",
      "options": [
        "Using all features every time",
        "Increasing tree depth",
        "Feature subsampling per tree",
        "Reducing number of trees"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Sampling a subset of features for each tree reduces overfitting and improves computation on high-dimensional data."
    },
    {
      "id": 25,
      "questionText": "Scenario: You want faster convergence with Gradient Boosting without losing accuracy. Strategy?",
      "options": [
        "Reduce tree depth to 1 always",
        "Increase learning rate drastically",
        "Lower learning rate slightly and increase n_estimators",
        "Use fewer features per tree only"
      ],
      "correctAnswerIndex": 2,
      "explanation": "A slightly lower learning rate combined with more estimators ensures stable, accurate learning while converging efficiently."
    },
    {
      "id": 26,
      "questionText": "Scenario: Your Gradient Boosting model is still overfitting after tuning learning rate. Next step?",
      "options": [
        "Use deeper trees",
        "Increase learning rate",
        "Reduce max_depth or min_samples_split",
        "Add more trees"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Controlling tree complexity by reducing depth or increasing minimum samples per split helps prevent overfitting."
    },
    {
      "id": 27,
      "questionText": "Which parameter limits the number of nodes in each tree?",
      "options": [
        "n_estimators",
        "max_leaf_nodes",
        "learning_rate",
        "subsample"
      ],
      "correctAnswerIndex": 1,
      "explanation": "max_leaf_nodes controls the maximum number of terminal nodes in each tree, limiting complexity."
    },
    {
      "id": 28,
      "questionText": "Scenario: You want to reduce variance without increasing bias in Gradient Boosting. Recommended action?",
      "options": [
        "Use only one deep tree",
        "Increase n_estimators and reduce learning rate",
        "Reduce number of features",
        "Increase learning rate significantly"
      ],
      "correctAnswerIndex": 1,
      "explanation": "More trees with lower learning rate reduce variance while preserving bias."
    },
    {
      "id": 29,
      "questionText": "What is the main difference between Stochastic Gradient Boosting and standard Gradient Boosting?",
      "options": [
        "Using deeper trees",
        "Subsampling of training data per tree",
        "Only one estimator is used",
        "Faster learning rate"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Stochastic Gradient Boosting trains each tree on a random subset of data, reducing variance and speeding training."
    },
    {
      "id": 30,
      "questionText": "Scenario: You are applying Gradient Boosting to a dataset with missing values. What is a standard approach?",
      "options": [
        "Use surrogate splits or imputation",
        "Remove all missing rows",
        "Use a single deep tree",
        "Ignore missing values"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Gradient Boosting can handle missing data using surrogate splits or by imputing values before training."
    },
    {
      "id": 31,
      "questionText": "Which metric can you monitor during Gradient Boosting training for early stopping?",
      "options": [
        "Validation loss or error",
        "Training set size",
        "Tree depth",
        "Number of features"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Monitoring validation loss allows early stopping to prevent overfitting while training."
    },
    {
      "id": 32,
      "questionText": "Scenario: You reduce learning rate but training becomes very slow. What is a good solution?",
      "options": [
        "Increase tree depth",
        "Stop training immediately",
        "Increase n_estimators to allow gradual learning",
        "Reduce dataset size drastically"
      ],
      "correctAnswerIndex": 2,
      "explanation": "A lower learning rate requires more trees (higher n_estimators) to fit the data effectively."
    },
    {
      "id": 33,
      "questionText": "What is the effect of increasing max_depth too much in Gradient Boosting?",
      "options": [
        "Reduction in variance",
        "Underfitting",
        "Faster convergence",
        "Overfitting"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Deep trees can model complex patterns but are prone to overfitting."
    },
    {
      "id": 34,
      "questionText": "Scenario: You use Gradient Boosting with very small n_estimators. Risk?",
      "options": [
        "Subsampling fails",
        "Learning rate becomes too high",
        "Overfitting immediately",
        "Underfitting due to insufficient model capacity"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Too few trees may prevent the model from capturing patterns in the data, leading to underfitting."
    },
    {
      "id": 35,
      "questionText": "Which of the following can help Gradient Boosting handle categorical variables?",
      "options": [
        "PCA",
        "Standard scaling only",
        "One-hot encoding or ordinal encoding",
        "Random subsampling"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Encoding categorical features allows Gradient Boosting trees to split effectively on categorical values."
    },
    {
      "id": 36,
      "questionText": "Scenario: Your model has slow convergence. Which combination is likely to improve it?",
      "options": [
        "Reduce subsample rate to 0.1",
        "Decrease learning rate and reduce trees",
        "Increase learning rate slightly and add more trees",
        "Reduce tree depth drastically only"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Slightly higher learning rate with more estimators can speed learning while maintaining accuracy."
    },
    {
      "id": 37,
      "questionText": "What is a common technique to prevent Gradient Boosting from overfitting noisy data?",
      "options": [
        "Remove subsampling",
        "Increase learning rate",
        "Increase tree depth",
        "Use shallow trees and lower learning rate"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Shallow trees and lower learning rate reduce the model's tendency to fit noise."
    },
    {
      "id": 38,
      "questionText": "Scenario: Using Gradient Boosting on imbalanced classes. Common adjustment?",
      "options": [
        "Use custom loss function or class weights",
        "Increase learning rate",
        "Ignore class imbalance",
        "Reduce n_estimators"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Weighted loss or custom loss functions help Gradient Boosting pay more attention to minority classes."
    },
    {
      "id": 39,
      "questionText": "Which technique allows Gradient Boosting to reduce correlation between trees?",
      "options": [
        "Using one tree only",
        "Reducing learning rate",
        "Increasing max_depth",
        "Subsampling data (stochastic boosting)"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Randomly sampling data for each tree reduces correlation, improving ensemble diversity and generalization."
    },
    {
      "id": 40,
      "questionText": "Scenario: You notice slow training with large n_estimators. Which option helps?",
      "options": [
        "Increase learning rate drastically",
        "Increase number of features per tree",
        "Remove subsampling",
        "Reduce max_depth or min_samples_split"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Simplifying trees reduces computation per estimator and speeds up training."
    },
    {
      "id": 41,
      "questionText": "Gradient Boosting sequentially adds trees to minimize which quantity?",
      "options": [
        "Feature variance",
        "Residual errors from previous trees",
        "Dataset size",
        "Learning rate"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Each new tree predicts the residual errors of the ensemble built so far."
    },
    {
      "id": 42,
      "questionText": "Scenario: Your model shows diminishing returns after many trees. Possible reason?",
      "options": [
        "Learning rate is zero",
        "Dataset is too large",
        "Residuals become small and difficult to improve",
        "Trees are too shallow"
      ],
      "correctAnswerIndex": 2,
      "explanation": "As the ensemble improves, residuals shrink, limiting the benefit of additional trees."
    },
    {
      "id": 43,
      "questionText": "Which variant of Gradient Boosting adapts to classification by optimizing logistic loss?",
      "options": [
        "Decision Tree Regression",
        "Stochastic Gradient Boosting",
        "AdaBoost",
        "Logistic Gradient Boosting"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Gradient Boosting can be adapted for classification by minimizing logistic loss."
    },
    {
      "id": 44,
      "questionText": "Scenario: Training Gradient Boosting on large dataset with limited memory. Strategy?",
      "options": [
        "Increase max_depth",
        "Increase learning rate",
        "Reduce subsample and feature fraction per tree",
        "Use full dataset each iteration"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Subsampling rows and features reduces memory usage and speeds up training."
    },
    {
      "id": 45,
      "questionText": "Which parameter controls how many features are used per tree in Gradient Boosting?",
      "options": [
        "max_features",
        "max_depth",
        "n_estimators",
        "learning_rate"
      ],
      "correctAnswerIndex": 0,
      "explanation": "max_features specifies the number of features considered for each tree, introducing randomness and reducing overfitting."
    },
    {
      "id": 46,
      "questionText": "Scenario: Your Gradient Boosting predictions are unstable. Likely cause?",
      "options": [
        "Low subsample",
        "Shallow trees",
        "High learning rate or deep trees",
        "Too few features"
      ],
      "correctAnswerIndex": 2,
      "explanation": "High learning rate and deep trees can cause the model to be sensitive to small data variations."
    },
    {
      "id": 47,
      "questionText": "Which type of problem is Gradient Boosting typically applied to?",
      "options": [
        "Clustering only",
        "Dimensionality reduction",
        "Regression and classification",
        "Feature extraction only"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Gradient Boosting is widely used for regression and classification tasks."
    },
    {
      "id": 48,
      "questionText": "Scenario: You want to combine Gradient Boosting with Random Forests. Benefit?",
      "options": [
        "Removes need for hyperparameter tuning",
        "Faster computation always",
        "Improved generalization by blending ensembles",
        "Reduces number of trees"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Blending ensembles can improve generalization but may not always reduce computation."
    },
    {
      "id": 49,
      "questionText": "What does the 'shrinkage' term refer to in Gradient Boosting?",
      "options": [
        "Learning rate",
        "Tree depth",
        "Number of features",
        "Subsample fraction"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Shrinkage is another term for the learning rate, controlling the contribution of each tree."
    },
    {
      "id": 50,
      "questionText": "Scenario: You increase subsample fraction to 1.0. Effect?",
      "options": [
        "Faster convergence always",
        "Reduces tree depth automatically",
        "Model underfits",
        "Less randomness, potentially higher overfitting"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Using the full dataset per iteration removes randomness and may increase overfitting."
    },
    {
      "id": 51,
      "questionText": "Scenario: Your Gradient Boosting model has high variance despite shallow trees. What could help?",
      "options": [
        "Use all features for each tree",
        "Increase learning rate",
        "Increase tree depth",
        "Reduce learning rate or use subsampling"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Reducing learning rate or using row/feature subsampling reduces variance and improves generalization."
    },
    {
      "id": 52,
      "questionText": "Which regularization technique is commonly applied in Gradient Boosting?",
      "options": [
        "Dropout",
        "Early stopping only",
        "L1/L2 penalties on leaf weights",
        "Batch normalization"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Some implementations (like XGBoost) allow L1/L2 regularization on leaf weights to prevent overfitting."
    },
    {
      "id": 53,
      "questionText": "Scenario: You want to reduce overfitting while keeping model complexity high. Best approach?",
      "options": [
        "Increase max_depth only",
        "Lower learning rate and increase n_estimators",
        "Increase learning rate",
        "Reduce subsample fraction to 0.1"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Lower learning rate with more trees allows high capacity without overfitting."
    },
    {
      "id": 54,
      "questionText": "What is the role of min_samples_split in Gradient Boosting trees?",
      "options": [
        "Learning rate",
        "Number of trees to build",
        "Maximum depth of tree",
        "Minimum number of samples required to split a node"
      ],
      "correctAnswerIndex": 3,
      "explanation": "min_samples_split controls the minimum samples needed to create a split, limiting overfitting."
    },
    {
      "id": 55,
      "questionText": "Scenario: You notice training is slow with very large n_estimators. Recommended action?",
      "options": [
        "Add more features",
        "Reduce number of trees",
        "Reduce max_depth or min_samples_split",
        "Increase learning rate drastically"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Simplifying trees reduces computation per estimator, speeding up training."
    },
    {
      "id": 56,
      "questionText": "Which loss function is commonly used for binary classification in Gradient Boosting?",
      "options": [
        "Mean squared error",
        "Euclidean distance",
        "Logistic loss (deviance)",
        "Huber loss"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Logistic loss is used to optimize Gradient Boosting for binary classification tasks."
    },
    {
      "id": 57,
      "questionText": "Scenario: You increase max_features to all features. Possible outcome?",
      "options": [
        "Higher risk of overfitting",
        "Less accurate predictions",
        "Faster training",
        "Reduced model capacity"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Using all features reduces randomness, which can increase overfitting."
    },
    {
      "id": 58,
      "questionText": "Which parameter controls the minimum number of samples in a leaf node?",
      "options": [
        "min_samples_leaf",
        "max_depth",
        "learning_rate",
        "n_estimators"
      ],
      "correctAnswerIndex": 0,
      "explanation": "min_samples_leaf prevents nodes with very few samples, reducing overfitting."
    },
    {
      "id": 59,
      "questionText": "Scenario: Your Gradient Boosting model struggles with high-dimensional sparse data. What helps?",
      "options": [
        "Increase learning rate",
        "Use fewer estimators",
        "Increase tree depth",
        "Feature subsampling per tree"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Subsampling features reduces complexity and improves generalization in high-dimensional sparse datasets."
    },
    {
      "id": 60,
      "questionText": "Which term describes sequentially fitting models to residual errors in Gradient Boosting?",
      "options": [
        "Feature scaling",
        "Bagging",
        "Random subsampling",
        "Gradient descent in function space"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Gradient Boosting performs gradient descent in function space by fitting new models to residuals."
    },
    {
      "id": 61,
      "questionText": "Scenario: You want Gradient Boosting to converge faster without overfitting. Strategy?",
      "options": [
        "Use fewer trees only",
        "Reduce max_depth to 1",
        "Slightly increase learning rate and add more trees",
        "Use very high learning rate"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Slightly higher learning rate with more estimators balances convergence speed and generalization."
    },
    {
      "id": 62,
      "questionText": "What does the subsample parameter control in stochastic Gradient Boosting?",
      "options": [
        "Learning rate",
        "Number of features per split",
        "Maximum depth",
        "Fraction of rows used per tree"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Subsample fraction determines how many training rows are randomly selected per iteration, introducing randomness."
    },
    {
      "id": 63,
      "questionText": "Scenario: You use Gradient Boosting for multiclass classification. Key adjustment?",
      "options": [
        "Use one-vs-rest or softmax loss",
        "Use mean squared error",
        "Use only one estimator",
        "Reduce tree depth to 1"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Multiclass problems require suitable loss functions or strategies like one-vs-rest or softmax."
    },
    {
      "id": 64,
      "questionText": "Which regularization parameter in XGBoost controls L2 penalty on leaf weights?",
      "options": [
        "gamma",
        "lambda",
        "alpha",
        "subsample"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Lambda applies L2 regularization to leaf weights, helping reduce overfitting."
    },
    {
      "id": 65,
      "questionText": "Scenario: You want to prevent Gradient Boosting from fitting noise in small datasets. Recommended?",
      "options": [
        "Increase learning rate",
        "Use all features per tree",
        "Increase max_depth drastically",
        "Lower learning rate and use shallow trees"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Lower learning rate and shallow trees reduce overfitting to noise."
    },
    {
      "id": 66,
      "questionText": "What is the effect of early stopping in Gradient Boosting?",
      "options": [
        "Increases learning rate",
        "Removes subsampling",
        "Stops training when validation loss stops improving",
        "Reduces tree depth automatically"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Early stopping prevents overfitting by halting training once performance on validation data plateaus."
    },
    {
      "id": 67,
      "questionText": "Scenario: You increase n_estimators and lower learning rate. Expected effect?",
      "options": [
        "Better generalization and lower bias",
        "Overfitting immediately",
        "Faster training only",
        "Model underfits always"
      ],
      "correctAnswerIndex": 0,
      "explanation": "More trees with smaller learning rate improves generalization while reducing bias."
    },
    {
      "id": 68,
      "questionText": "Which Gradient Boosting variant uses row and feature subsampling?",
      "options": [
        "AdaBoost",
        "Bagging",
        "Standard Gradient Boosting",
        "Stochastic Gradient Boosting"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Stochastic Gradient Boosting introduces randomness by subsampling rows and/or features per tree."
    },
    {
      "id": 69,
      "questionText": "Scenario: Your Gradient Boosting model predicts extreme values for outliers. Solution?",
      "options": [
        "Increase tree depth",
        "Increase learning rate",
        "Remove subsampling",
        "Use robust loss function like Huber loss"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Robust loss functions reduce sensitivity to outliers."
    },
    {
      "id": 70,
      "questionText": "What does gamma (min_split_loss) control in XGBoost?",
      "options": [
        "Maximum depth",
        "Number of estimators",
        "Learning rate",
        "Minimum loss reduction required to make a split"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Gamma prevents unnecessary splits by requiring a minimum loss reduction for a node to split."
    },
    {
      "id": 71,
      "questionText": "Scenario: Using Gradient Boosting for regression. Best practice?",
      "options": [
        "Increase tree depth aggressively",
        "Use only one deep tree",
        "Ignore validation set",
        "Monitor validation loss and adjust learning rate/n_estimators"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Monitoring validation loss ensures good generalization and proper parameter tuning."
    },
    {
      "id": 72,
      "questionText": "Which technique can reduce Gradient Boosting training time on large datasets?",
      "options": [
        "Increase learning rate drastically",
        "Add more estimators",
        "Use more features per tree",
        "Subsample rows and features, limit tree depth"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Row/feature subsampling and shallow trees reduce computation and memory usage."
    },
    {
      "id": 73,
      "questionText": "Scenario: Gradient Boosting produces unstable predictions. Likely cause?",
      "options": [
        "Shallow trees",
        "Low subsample fraction",
        "High learning rate or deep trees",
        "Low n_estimators"
      ],
      "correctAnswerIndex": 2,
      "explanation": "High learning rate and very deep trees can make predictions sensitive to small variations in data."
    },
    {
      "id": 74,
      "questionText": "Which ensemble technique is Gradient Boosting based on?",
      "options": [
        "Boosting",
        "Voting",
        "Stacking",
        "Bagging"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Gradient Boosting is a boosting technique, sequentially correcting errors of weak learners."
    },
    {
      "id": 75,
      "questionText": "Scenario: You want Gradient Boosting to generalize better on small dataset. Effective approach?",
      "options": [
        "Increase max_depth only",
        "Use all features per tree without subsampling",
        "Lower learning rate, reduce tree depth, use subsampling",
        "Increase learning rate drastically"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Reducing learning rate, limiting tree depth, and using subsampling helps prevent overfitting on small datasets."
    },
    {
      "id": 76,
      "questionText": "Scenario: Gradient Boosting predictions fluctuate between runs. Likely cause?",
      "options": [
        "Shallow trees",
        "Early stopping",
        "High learning rate or no subsampling",
        "Too few features"
      ],
      "correctAnswerIndex": 2,
      "explanation": "High learning rate and lack of subsampling can make predictions unstable across different runs."
    },
    {
      "id": 77,
      "questionText": "Which parameter can help prevent Gradient Boosting from creating overly complex trees?",
      "options": [
        "max_depth",
        "learning_rate",
        "subsample",
        "n_estimators"
      ],
      "correctAnswerIndex": 0,
      "explanation": "max_depth limits the maximum depth of individual trees, controlling complexity."
    },
    {
      "id": 78,
      "questionText": "Scenario: Model overfits training data despite tuning learning rate and n_estimators. Additional fix?",
      "options": [
        "Reduce max_depth or increase min_samples_leaf",
        "Increase learning rate",
        "Increase max_features to all",
        "Use fewer trees"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Controlling tree complexity with max_depth or min_samples_leaf helps reduce overfitting."
    },
    {
      "id": 79,
      "questionText": "Which variant of Gradient Boosting introduces randomness by subsampling both rows and features?",
      "options": [
        "Bagging",
        "AdaBoost",
        "Standard Gradient Boosting",
        "Stochastic Gradient Boosting"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Stochastic Gradient Boosting uses row and feature subsampling per tree to improve generalization."
    },
    {
      "id": 80,
      "questionText": "Scenario: Using Gradient Boosting with noisy data. Best practice?",
      "options": [
        "Lower learning rate, shallow trees, possibly subsample rows",
        "Increase learning rate",
        "Use all features per tree",
        "Use very deep trees"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Shallow trees, lower learning rate, and subsampling reduce overfitting to noise."
    },
    {
      "id": 81,
      "questionText": "What does min_samples_split control in Gradient Boosting?",
      "options": [
        "Number of estimators",
        "Maximum depth of trees",
        "Learning rate",
        "Minimum samples required to split a node"
      ],
      "correctAnswerIndex": 3,
      "explanation": "min_samples_split prevents splitting nodes with very few samples, helping reduce overfitting."
    },
    {
      "id": 82,
      "questionText": "Scenario: Validation loss increases after several iterations. Solution?",
      "options": [
        "Apply early stopping",
        "Use deeper trees",
        "Add more trees regardless",
        "Increase learning rate"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Early stopping halts training when validation loss stops improving to prevent overfitting."
    },
    {
      "id": 83,
      "questionText": "Which parameter scales the contribution of each tree in Gradient Boosting?",
      "options": [
        "max_depth",
        "subsample",
        "learning_rate (shrinkage)",
        "n_estimators"
      ],
      "correctAnswerIndex": 2,
      "explanation": "The learning_rate (shrinkage) controls how much each tree contributes to the ensemble."
    },
    {
      "id": 84,
      "questionText": "Scenario: Model is slow on large dataset. Best strategies?",
      "options": [
        "Increase learning rate drastically",
        "Use more features per tree",
        "Reduce max_depth, min_samples_split, or use subsampling",
        "Reduce learning rate to zero"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Simplifying trees and using subsampling reduces computation and memory usage."
    },
    {
      "id": 85,
      "questionText": "Which loss function is used for multi-class classification in Gradient Boosting?",
      "options": [
        "Mean squared error",
        "Softmax / multinomial deviance",
        "Huber loss",
        "Binary cross-entropy"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Softmax or multinomial deviance loss is used for multi-class classification problems."
    },
    {
      "id": 86,
      "questionText": "Scenario: Gradient Boosting underfits. What adjustment helps?",
      "options": [
        "Use fewer features",
        "Increase tree depth or n_estimators",
        "Apply early stopping immediately",
        "Reduce learning rate drastically"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Increasing tree depth or number of estimators allows the model to better fit the data."
    },
    {
      "id": 87,
      "questionText": "Which regularization parameter in XGBoost applies L1 penalty on leaf weights?",
      "options": [
        "gamma",
        "lambda",
        "subsample",
        "alpha"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Alpha applies L1 regularization to leaf weights, helping prevent overfitting."
    },
    {
      "id": 88,
      "questionText": "Scenario: Learning rate is low and n_estimators are small. Risk?",
      "options": [
        "Noise sensitivity",
        "Overfitting",
        "Random predictions",
        "Underfitting"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Low learning rate with few trees prevents the model from fitting patterns, leading to underfitting."
    },
    {
      "id": 89,
      "questionText": "Scenario: You increase subsample to 1.0. Effect?",
      "options": [
        "Less randomness and higher risk of overfitting",
        "Faster convergence",
        "Underfitting",
        "Reduced tree depth"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Using all data removes randomness and can increase overfitting."
    },
    {
      "id": 90,
      "questionText": "Which technique reduces correlation among Gradient Boosting trees?",
      "options": [
        "Increasing max_depth",
        "Increasing learning rate",
        "Row and feature subsampling",
        "Using single tree"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Random sampling of rows and features reduces correlation between trees and improves generalization."
    },
    {
      "id": 91,
      "questionText": "Scenario: Validation performance plateaus before n_estimators. Recommended?",
      "options": [
        "Increase learning rate drastically",
        "Add more features",
        "Use early stopping",
        "Increase max_depth"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Early stopping halts training when validation performance stops improving to avoid overfitting."
    },
    {
      "id": 92,
      "questionText": "Scenario: Predictions are too sensitive to outliers. Solution?",
      "options": [
        "Reduce n_estimators",
        "Increase learning rate",
        "Use deeper trees",
        "Use robust loss function like Huber loss"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Robust loss functions reduce sensitivity to extreme values."
    },
    {
      "id": 93,
      "questionText": "Which Gradient Boosting implementation allows L1/L2 regularization and parallelization?",
      "options": [
        "AdaBoost",
        "XGBoost",
        "Bagging",
        "Scikit-learn GradientBoosting"
      ],
      "correctAnswerIndex": 1,
      "explanation": "XGBoost supports advanced regularization and parallel computation."
    },
    {
      "id": 94,
      "questionText": "Scenario: You want Gradient Boosting to generalize on high-dimensional sparse data. Approach?",
      "options": [
        "Increase tree depth",
        "Use all rows always",
        "Increase learning rate",
        "Subsample features per tree"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Feature subsampling reduces complexity and overfitting in sparse, high-dimensional data."
    },
    {
      "id": 95,
      "questionText": "Scenario: Model predicts extreme residuals for outliers. Solution?",
      "options": [
        "Increase max_depth",
        "Reduce subsample",
        "Use robust loss function",
        "Increase learning rate"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Robust loss functions like Huber loss reduce influence of outliers."
    },
    {
      "id": 96,
      "questionText": "Which parameter controls minimum loss reduction required to make a split in XGBoost?",
      "options": [
        "gamma (min_split_loss)",
        "alpha",
        "lambda",
        "subsample"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Gamma prevents splits that do not improve the loss function sufficiently."
    },
    {
      "id": 97,
      "questionText": "Scenario: Small dataset shows overfitting. Strategy?",
      "options": [
        "Use all features per tree",
        "Increase learning rate",
        "Increase max_depth",
        "Reduce learning rate, shallow trees, use subsampling"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Lower learning rate, shallow trees, and subsampling help prevent overfitting on small datasets."
    },
    {
      "id": 98,
      "questionText": "Which ensemble method is Gradient Boosting part of?",
      "options": [
        "Stacking",
        "Bagging",
        "Voting",
        "Boosting"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Gradient Boosting is a boosting method, combining weak learners sequentially to reduce error."
    },
    {
      "id": 99,
      "questionText": "Scenario: High variance despite using shallow trees and low learning rate. Possible fix?",
      "options": [
        "Increase tree depth",
        "Increase subsample fraction and feature randomness",
        "Increase learning rate",
        "Reduce number of estimators"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Subsampling rows and features introduces randomness and reduces variance."
    },
    {
      "id": 100,
      "questionText": "Scenario: You need Gradient Boosting to handle multiclass classification efficiently. Best approach?",
      "options": [
        "Use mean squared error",
        "Ignore class differences",
        "Use softmax/multinomial loss with suitable n_estimators and learning rate",
        "Use a single tree per class"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Softmax/multinomial loss allows proper multiclass classification with Gradient Boosting."
    }
  ]
}