{
  "title": "Support Vector Machines (SVM) Mastery: 100 MCQs",
  "description": "A complete 100-question collection designed to teach and test your understanding of Support Vector Machines — from basic margin intuition to advanced kernel tricks, soft margin optimization, hyperparameter tuning, and real-world scenario applications.",
  "questions": [
    {
      "id": 1,
      "questionText": "What does an SVM aim to find in the feature space?",
      "options": [
        "A random boundary",
        "A centroid of all data points",
        "A hyperplane that maximizes margin",
        "A cluster center"
      ],
      "correctAnswerIndex": 2,
      "explanation": "SVM aims to find the optimal separating hyperplane that maximizes the margin between classes."
    },
    {
      "id": 2,
      "questionText": "Scenario: SVM is trained on perfectly separable data. Which margin type is used?",
      "options": [
        "No margin",
        "Random margin",
        "Soft margin",
        "Hard margin"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Hard margin SVM is used when data is perfectly linearly separable."
    },
    {
      "id": 3,
      "questionText": "Scenario: Data contains overlapping classes. Which SVM variation should be used?",
      "options": [
        "Decision trees",
        "Hard margin",
        "Soft margin",
        "Polynomial kernel only"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Soft margin SVM allows some misclassification to handle overlapping data."
    },
    {
      "id": 4,
      "questionText": "What is the primary role of support vectors?",
      "options": [
        "Maximize dataset size",
        "Define the decision boundary",
        "Increase margin penalty",
        "Reduce dimensions"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Support vectors are the critical points that define the position and orientation of the separating hyperplane."
    },
    {
      "id": 5,
      "questionText": "Scenario: Linear SVM trained on non-linear data. What is likely?",
      "options": [
        "Perfect accuracy",
        "Underfitting occurs",
        "Zero training loss",
        "Overfitting occurs"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Linear SVMs cannot model non-linear relationships, leading to underfitting."
    },
    {
      "id": 6,
      "questionText": "Which kernel function maps data to infinite-dimensional space?",
      "options": [
        "Linear",
        "RBF (Gaussian)",
        "Polynomial",
        "Sigmoid"
      ],
      "correctAnswerIndex": 1,
      "explanation": "The RBF kernel maps data into an infinite-dimensional feature space, enabling complex boundaries."
    },
    {
      "id": 7,
      "questionText": "Scenario: SVM with RBF kernel and γ is too large. Effect?",
      "options": [
        "Acts like linear",
        "Overfits training data",
        "Fails to converge",
        "Underfits"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Large γ makes the model focus too much on each point, overfitting the training set."
    },
    {
      "id": 8,
      "questionText": "Scenario: SVM trained with small C. What happens?",
      "options": [
        "Overfits training data",
        "Allows more misclassifications",
        "Creates zero margin",
        "Fails to train"
      ],
      "correctAnswerIndex": 1,
      "explanation": "A smaller C allows wider margins and tolerates more errors for better generalization."
    },
    {
      "id": 9,
      "questionText": "Scenario: Large C used with noisy data. Effect?",
      "options": [
        "Reduces kernel complexity",
        "Ignores outliers",
        "Overfits noise",
        "Underfits"
      ],
      "correctAnswerIndex": 2,
      "explanation": "A large C emphasizes classification accuracy, possibly overfitting noisy samples."
    },
    {
      "id": 10,
      "questionText": "Why is feature scaling critical for SVM?",
      "options": [
        "To remove duplicates",
        "To normalize labels",
        "Because SVM depends on distance calculations",
        "To convert categorical data"
      ],
      "correctAnswerIndex": 2,
      "explanation": "SVM uses dot products and distance metrics; scaling prevents feature dominance."
    },
    {
      "id": 11,
      "questionText": "Scenario: Two features have vastly different ranges. What happens if not scaled?",
      "options": [
        "No impact",
        "Better accuracy",
        "Model bias towards larger scale feature",
        "Faster convergence"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Unscaled features distort margin calculations, biasing the model."
    },
    {
      "id": 12,
      "questionText": "What is the role of the kernel trick?",
      "options": [
        "Reduces features",
        "Improves feature scaling",
        "Maps data to higher dimensions without explicit transformation",
        "Normalizes data"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Kernel trick lets SVM handle non-linear data efficiently without explicit transformation."
    },
    {
      "id": 13,
      "questionText": "Scenario: SVM applied to high-dimensional text data. Best kernel?",
      "options": [
        "Sigmoid kernel",
        "Linear kernel",
        "RBF kernel",
        "Polynomial kernel"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Linear SVMs perform well for high-dimensional sparse data such as text."
    },
    {
      "id": 14,
      "questionText": "Scenario: Non-linear boundaries observed. Which kernel is best?",
      "options": [
        "No kernel",
        "RBF kernel",
        "Sigmoid kernel only",
        "Linear kernel"
      ],
      "correctAnswerIndex": 1,
      "explanation": "The RBF kernel can model highly non-linear decision boundaries."
    },
    {
      "id": 15,
      "questionText": "What does γ control in an RBF kernel?",
      "options": [
        "Regularization strength",
        "The influence of a single training example",
        "Learning rate",
        "Loss function type"
      ],
      "correctAnswerIndex": 1,
      "explanation": "γ defines how far the influence of a training sample reaches; higher γ = closer reach."
    },
    {
      "id": 16,
      "questionText": "Scenario: γ too small in RBF kernel. Effect?",
      "options": [
        "Zero accuracy",
        "Underfits; boundary too smooth",
        "Fails to converge",
        "Overfits"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Small γ makes the model too smooth, underfitting complex patterns."
    },
    {
      "id": 17,
      "questionText": "What happens if C=∞ in soft-margin SVM?",
      "options": [
        "Ignores support vectors",
        "Always fails",
        "Creates random margins",
        "Behaves like hard-margin SVM"
      ],
      "correctAnswerIndex": 3,
      "explanation": "When C is very large, SVM tries to classify all points correctly like a hard-margin model."
    },
    {
      "id": 18,
      "questionText": "Scenario: SVM used for regression (SVR). What is optimized?",
      "options": [
        "Epsilon-insensitive loss",
        "Huber loss",
        "Cross-entropy",
        "Hinge loss"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Support Vector Regression uses epsilon-insensitive loss for fitting continuous data."
    },
    {
      "id": 19,
      "questionText": "What happens if all points lie outside the margin in SVM?",
      "options": [
        "Margin expands",
        "Kernel fails",
        "Model complexity increases",
        "C ignored"
      ],
      "correctAnswerIndex": 2,
      "explanation": "If most points lie outside, the penalty term increases model complexity."
    },
    {
      "id": 20,
      "questionText": "Scenario: SVM trained with too many features but few samples. Risk?",
      "options": [
        "Overfitting",
        "Perfect generalization",
        "Fast convergence",
        "Underfitting"
      ],
      "correctAnswerIndex": 0,
      "explanation": "High feature-to-sample ratio leads to overfitting."
    },
    {
      "id": 21,
      "questionText": "What does the bias term in SVM represent?",
      "options": [
        "C penalty",
        "The offset of the hyperplane",
        "Learning rate",
        "The variance"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Bias determines how far the decision boundary is from the origin."
    },
    {
      "id": 22,
      "questionText": "Scenario: RBF kernel with optimal γ and large C. Expected result?",
      "options": [
        "Linear decision boundary",
        "Overfit training set",
        "Underfit",
        "Ignore support vectors"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Large C and high γ both risk overfitting due to complex boundaries."
    },
    {
      "id": 23,
      "questionText": "Why does SVM not work well with large datasets?",
      "options": [
        "Cannot handle linear data",
        "Training time increases quadratically",
        "Too few features",
        "Memory always freed"
      ],
      "correctAnswerIndex": 1,
      "explanation": "SVM training complexity scales poorly with data size."
    },
    {
      "id": 24,
      "questionText": "Scenario: SVM applied with polynomial kernel degree=10. What happens?",
      "options": [
        "Overfits data",
        "Linear boundary",
        "Underfits",
        "No effect"
      ],
      "correctAnswerIndex": 0,
      "explanation": "High-degree polynomial kernels can easily overfit."
    },
    {
      "id": 25,
      "questionText": "What is hinge loss used for?",
      "options": [
        "Hyperparameter tuning",
        "Feature selection",
        "Margin-based classification",
        "Regression"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Hinge loss is used in SVM to measure margin violations."
    },
    {
      "id": 26,
      "questionText": "Scenario: Noisy dataset with overlapping features. Best SVM approach?",
      "options": [
        "Linear only",
        "Soft margin with small C",
        "High γ",
        "Hard margin"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Soft margin and smaller C improve tolerance to noise."
    },
    {
      "id": 27,
      "questionText": "Scenario: Model overfits using RBF kernel. Possible fix?",
      "options": [
        "Remove regularization",
        "Reduce γ",
        "Increase γ",
        "Increase C"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Reducing γ smooths decision boundaries to avoid overfitting."
    },
    {
      "id": 28,
      "questionText": "Scenario: Data not linearly separable but low-dimensional. Efficient kernel?",
      "options": [
        "Polynomial kernel (degree 2)",
        "Linear",
        "RBF kernel",
        "Sigmoid"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Low-degree polynomial kernels can model slight non-linearities efficiently."
    },
    {
      "id": 29,
      "questionText": "Scenario: You use an RBF kernel on data with high dimensionality and little noise. What might happen?",
      "options": [
        "Good fit if parameters tuned",
        "Ignores all kernels",
        "Underfits automatically",
        "Always overfits"
      ],
      "correctAnswerIndex": 0,
      "explanation": "High-dimensional data can work well with RBF kernels if hyperparameters are well-tuned."
    },
    {
      "id": 30,
      "questionText": "Scenario: Polynomial kernel used with degree=1. What kernel does this mimic?",
      "options": [
        "RBF kernel",
        "No kernel",
        "Sigmoid kernel",
        "Linear kernel"
      ],
      "correctAnswerIndex": 3,
      "explanation": "A polynomial kernel with degree 1 is equivalent to a linear kernel."
    },
    {
      "id": 31,
      "questionText": "Scenario: γ in RBF kernel set to 0.001. What happens?",
      "options": [
        "Acts as linear kernel",
        "Fails to converge",
        "Model overfits",
        "Model underfits; boundary too smooth"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Very small γ makes the RBF behave almost linearly, leading to underfitting."
    },
    {
      "id": 32,
      "questionText": "Scenario: Multiclass classification with SVM. Which strategy is used?",
      "options": [
        "One-vs-Rest or One-vs-One",
        "Naive Bayes",
        "K-Means",
        "Softmax"
      ],
      "correctAnswerIndex": 0,
      "explanation": "SVM handles multiclass via one-vs-rest or one-vs-one strategies."
    },
    {
      "id": 33,
      "questionText": "Scenario: SVM trained on imbalanced data. What may occur?",
      "options": [
        "Perfect accuracy",
        "Bias toward majority class",
        "Bias toward minority",
        "Uniform decision boundary"
      ],
      "correctAnswerIndex": 1,
      "explanation": "SVM may favor the majority class unless class weights are balanced."
    },
    {
      "id": 34,
      "questionText": "How does SVM handle non-linear separations?",
      "options": [
        "By increasing epochs",
        "By removing bias",
        "By adding dropout",
        "By using kernel functions"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Kernels allow SVMs to map data into higher-dimensional spaces to handle non-linearity."
    },
    {
      "id": 35,
      "questionText": "Scenario: Large C and large γ chosen for RBF kernel. Expected behavior?",
      "options": [
        "Overfitting",
        "Stable model",
        "Underfitting",
        "Fails to train"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Large C and γ can both cause complex decision boundaries, leading to overfitting."
    },
    {
      "id": 36,
      "questionText": "Scenario: You increase C from 1 to 1000. What happens?",
      "options": [
        "Margin widens",
        "Kernel ignored",
        "Margin becomes narrower",
        "Model underfits"
      ],
      "correctAnswerIndex": 2,
      "explanation": "A larger C penalizes misclassifications more, resulting in a narrower margin."
    },
    {
      "id": 37,
      "questionText": "Which optimization technique does SVM use to find the best hyperplane?",
      "options": [
        "Simulated annealing",
        "Gradient descent",
        "Quadratic programming",
        "Stochastic optimization"
      ],
      "correctAnswerIndex": 2,
      "explanation": "SVMs use quadratic programming to solve the optimization problem."
    },
    {
      "id": 38,
      "questionText": "Scenario: High γ, low C combination. Expected result?",
      "options": [
        "Complex boundary but tolerates errors",
        "Training failure",
        "Linear separation",
        "Perfect fit"
      ],
      "correctAnswerIndex": 0,
      "explanation": "High γ adds complexity, but low C softens penalties, balancing flexibility."
    },
    {
      "id": 39,
      "questionText": "Scenario: SVM fails to converge. Possible reason?",
      "options": [
        "Improper scaling or large C/γ",
        "Too few features",
        "Low variance",
        "Kernel not imported"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Unscaled data or extreme parameter values can cause convergence issues."
    },
    {
      "id": 40,
      "questionText": "Why does SVM perform well in high-dimensional spaces?",
      "options": [
        "It uses PCA internally",
        "It ignores most features",
        "It compresses data",
        "It depends on support vectors, not dimensionality"
      ],
      "correctAnswerIndex": 3,
      "explanation": "SVM focuses on boundary points (support vectors), not the entire space."
    },
    {
      "id": 41,
      "questionText": "Scenario: Features highly correlated. Impact on SVM?",
      "options": [
        "Minimal impact; still works well",
        "Fails to classify",
        "Reduces C",
        "Doubles training time"
      ],
      "correctAnswerIndex": 0,
      "explanation": "SVMs can still work well but may benefit from decorrelation or PCA."
    },
    {
      "id": 42,
      "questionText": "Scenario: RBF kernel underfits training data. Possible fix?",
      "options": [
        "Use linear kernel",
        "Remove kernel",
        "Decrease γ",
        "Increase γ or C"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Higher γ or C increases flexibility and reduces underfitting."
    },
    {
      "id": 43,
      "questionText": "Scenario: Linear kernel chosen for non-linear data. Expected result?",
      "options": [
        "Balanced model",
        "Underfitting",
        "Perfect fit",
        "Overfitting"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Linear kernels cannot capture complex patterns, leading to underfitting."
    },
    {
      "id": 44,
      "questionText": "What happens if all data points are support vectors?",
      "options": [
        "Underfitting",
        "Overfitting",
        "No change",
        "Perfect generalization"
      ],
      "correctAnswerIndex": 1,
      "explanation": "If every point influences the boundary, the model likely overfits."
    },
    {
      "id": 45,
      "questionText": "Scenario: You observe slow SVM training on large dataset. What can help?",
      "options": [
        "Add regularization",
        "Increase γ",
        "Use linear SVM (LinearSVC)",
        "Use higher-degree kernel"
      ],
      "correctAnswerIndex": 2,
      "explanation": "LinearSVC is optimized for large-scale linear problems."
    },
    {
      "id": 46,
      "questionText": "Scenario: Dataset has noise and outliers. Which SVM parameter to tune?",
      "options": [
        "C (regularization)",
        "Degree",
        "γ",
        "Bias"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Smaller C helps tolerate misclassifications and handle noisy data."
    },
    {
      "id": 47,
      "questionText": "Scenario: SVM used on normalized image features. Kernel to start with?",
      "options": [
        "Polynomial (deg=3)",
        "RBF",
        "Linear",
        "Sigmoid"
      ],
      "correctAnswerIndex": 1,
      "explanation": "RBF kernel often performs well on normalized, moderate-dimensional data."
    },
    {
      "id": 48,
      "questionText": "Scenario: γ=0.0001 and C=1000. Likely effect?",
      "options": [
        "Overfitting",
        "No convergence",
        "Underfitting",
        "Optimal fit"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Extremely low γ makes decision boundary too simple, underfitting occurs."
    },
    {
      "id": 49,
      "questionText": "Scenario: Multiclass SVM classification accuracy drops. Fix?",
      "options": [
        "Add dropout",
        "Use balanced class weights",
        "Reduce features",
        "Switch to regression"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Balancing class weights helps when class imbalance causes bias."
    },
    {
      "id": 50,
      "questionText": "Scenario: You test SVM on polynomial kernel degree=6. Observation?",
      "options": [
        "Underfits large datasets",
        "Ignores bias",
        "Linearizes output",
        "Overfits small datasets"
      ],
      "correctAnswerIndex": 3,
      "explanation": "High-degree polynomial kernels often overfit, especially on limited data."
    },
    {
      "id": 51,
      "questionText": "Scenario: SVM trained with sigmoid kernel. What does it resemble?",
      "options": [
        "RBF mapping",
        "Decision tree splitting",
        "Linear regression",
        "Neural network activation function"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Sigmoid kernel mimics a neural network activation behavior."
    },
    {
      "id": 52,
      "questionText": "What is the dual formulation used for in SVM?",
      "options": [
        "To reduce memory usage",
        "To handle high-dimensional kernels",
        "To remove bias term",
        "To normalize outputs"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Dual formulation helps apply kernel trick efficiently in high-dimensional space."
    },
    {
      "id": 53,
      "questionText": "Scenario: SVM used for spam detection (text data). Best kernel?",
      "options": [
        "Polynomial",
        "RBF",
        "Sigmoid",
        "Linear"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Linear kernels are efficient and effective for sparse text data."
    },
    {
      "id": 54,
      "questionText": "Scenario: Overfitting in SVM model. Which parameter should be reduced?",
      "options": [
        "Loss function",
        "C or γ",
        "Degree",
        "Bias"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Reducing C or γ simplifies the model and improves generalization."
    },
    {
      "id": 55,
      "questionText": "Scenario: Dataset has millions of samples. SVM alternative?",
      "options": [
        "Sigmoid SVM",
        "Stochastic gradient linear classifier",
        "Polynomial SVM",
        "Decision tree"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Large datasets often use SGD-based linear classifiers for scalability."
    },
    {
      "id": 56,
      "questionText": "Scenario: You use a kernel not positive semi-definite. What may occur?",
      "options": [
        "Better accuracy",
        "Optimization fails",
        "Underfitting",
        "Overfitting"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Non-PSD kernels can violate convex optimization requirements."
    },
    {
      "id": 57,
      "questionText": "Scenario: SVM applied for anomaly detection. Variant used?",
      "options": [
        "One-Class SVM",
        "K-Means",
        "SVR",
        "Binary SVM"
      ],
      "correctAnswerIndex": 0,
      "explanation": "One-Class SVM is designed for novelty or anomaly detection tasks."
    },
    {
      "id": 58,
      "questionText": "Scenario: Data contains many irrelevant features. Approach?",
      "options": [
        "Lower γ",
        "Feature selection before SVM",
        "Add more kernels",
        "Increase C"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Feature selection reduces noise and improves SVM performance."
    },
    {
      "id": 59,
      "questionText": "Scenario: SVM decision boundary oscillates too much. Cause?",
      "options": [
        "Linear kernel",
        "Large γ",
        "Small C",
        "Small γ"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Large γ makes decision boundaries sensitive to individual samples."
    },
    {
      "id": 60,
      "questionText": "Scenario: You tune γ=0.1, C=10 via grid search. Effect?",
      "options": [
        "Guaranteed overfit",
        "Always underfit",
        "Improved generalization",
        "No change"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Grid search helps find the optimal trade-off between bias and variance."
    },
    {
      "id": 61,
      "questionText": "Scenario: SVM used with PCA-transformed features. Benefit?",
      "options": [
        "Faster convergence and less overfitting",
        "No benefit",
        "Worse accuracy",
        "Kernel ignored"
      ],
      "correctAnswerIndex": 0,
      "explanation": "PCA reduces redundancy, improving SVM performance and speed."
    },
    {
      "id": 62,
      "questionText": "Scenario: RBF kernel accuracy drops on test data. Likely reason?",
      "options": [
        "Kernel removed",
        "Overfitting due to high γ",
        "Underfitting",
        "Noisy training data"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Too high γ causes overfitting, reducing test performance."
    },
    {
      "id": 63,
      "questionText": "Scenario: SVM predicts continuous target. Variant?",
      "options": [
        "Kernel Ridge",
        "SVR (Support Vector Regression)",
        "Soft-margin SVM",
        "Linear SVM"
      ],
      "correctAnswerIndex": 1,
      "explanation": "SVR adapts the SVM principle for regression tasks."
    },
    {
      "id": 64,
      "questionText": "Scenario: You combine linear and RBF kernels. Effect?",
      "options": [
        "Error increases",
        "Kernel ignored",
        "Hybrid decision surface",
        "No benefit"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Hybrid kernels can model both linear and non-linear relationships."
    },
    {
      "id": 65,
      "questionText": "Scenario: SVM model gives different results on same data. Cause?",
      "options": [
        "Kernel mismatch",
        "Scaling issue",
        "Different γ",
        "Non-deterministic solver or random state"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Different random seeds or solvers can yield slightly varying solutions."
    },
    {
      "id": 66,
      "questionText": "Scenario: Class imbalance severe (90:10). Recommended?",
      "options": [
        "Use sigmoid kernel",
        "Reduce features",
        "Increase C",
        "Use class_weight='balanced'"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Setting class_weight='balanced' compensates for imbalance."
    },
    {
      "id": 67,
      "questionText": "Scenario: SVM on dataset with 10M samples. Efficient library?",
      "options": [
        "Polynomial kernel",
        "Naive Bayes",
        "RBF kernel SVC",
        "LinearSVC or SGDClassifier"
      ],
      "correctAnswerIndex": 3,
      "explanation": "LinearSVC or SGDClassifier scale better for large data."
    },
    {
      "id": 68,
      "questionText": "Scenario: High variance SVM results. Remedy?",
      "options": [
        "Increase γ",
        "Use hard margin",
        "Use cross-validation and parameter tuning",
        "Add noise"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Cross-validation stabilizes and selects optimal hyperparameters."
    },
    {
      "id": 69,
      "questionText": "Scenario: Feature scaling forgotten before training. Effect?",
      "options": [
        "Incorrect margin calculation",
        "Higher recall",
        "Faster training",
        "Better accuracy"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Unscaled data distorts distance-based calculations in SVM."
    },
    {
      "id": 70,
      "questionText": "Scenario: SVM trained with kernel='poly', degree=5. What to expect?",
      "options": [
        "No margin",
        "Underfitting",
        "Overfitting",
        "Perfect generalization"
      ],
      "correctAnswerIndex": 2,
      "explanation": "High-degree polynomial kernels tend to overfit."
    },
    {
      "id": 71,
      "questionText": "Scenario: You visualize decision boundary very smooth. Cause?",
      "options": [
        "High degree",
        "Small C",
        "Small γ",
        "Large γ"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Small γ creates smoother, less complex boundaries."
    },
    {
      "id": 72,
      "questionText": "Scenario: You want probabilistic outputs from SVM. How?",
      "options": [
        "Enable probability=True",
        "Use RBF kernel",
        "Disable scaling",
        "Increase C"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Enabling probability=True uses Platt scaling to estimate probabilities."
    },
    {
      "id": 73,
      "questionText": "Scenario: Training time too high with kernel SVM. Remedy?",
      "options": [
        "Use linear approximation",
        "Add noise",
        "Increase C",
        "Increase γ"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Approximation methods like LinearSVC or kernel approximation speed up training."
    },
    {
      "id": 74,
      "questionText": "Scenario: SVM applied on binary imbalanced medical dataset. Recommendation?",
      "options": [
        "Drop small class",
        "Hard margin",
        "Random validation",
        "Stratified cross-validation"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Stratified cross-validation preserves class ratio during validation."
    },
    {
      "id": 75,
      "questionText": "Scenario: Model accuracy high on train, low on test. Issue?",
      "options": [
        "Low variance",
        "Overfitting",
        "Scaling error",
        "Underfitting"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Overfitting occurs when SVM learns noise and fails to generalize."
    },
    {
      "id": 76,
      "questionText": "Scenario: Feature correlation high, training unstable. Fix?",
      "options": [
        "Apply PCA before SVM",
        "Use high γ",
        "Reduce support vectors",
        "Increase C"
      ],
      "correctAnswerIndex": 0,
      "explanation": "PCA helps reduce correlation and stabilize training."
    },
    {
      "id": 77,
      "questionText": "Scenario: You want to visualize margin width. Which SVM attribute?",
      "options": [
        "kernel_",
        "n_iter_",
        "coef_ and intercept_",
        "support_"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Margin width can be computed using coef_ and intercept_ in linear SVM."
    },
    {
      "id": 78,
      "questionText": "Scenario: You use polynomial kernel degree=3. What’s effect?",
      "options": [
        "Fail to train",
        "Linear separation",
        "Non-linear decision surface",
        "Underfitting"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Polynomial kernel allows curved decision boundaries."
    },
    {
      "id": 79,
      "questionText": "Scenario: SVM used for text sentiment analysis. Kernel?",
      "options": [
        "Sigmoid",
        "Linear",
        "RBF",
        "Polynomial"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Linear kernel works best for high-dimensional sparse text data."
    },
    {
      "id": 80,
      "questionText": "Scenario: Decision boundary too sensitive to single points. Cause?",
      "options": [
        "Balanced class weights",
        "Small γ",
        "Large γ",
        "Small C"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Large γ focuses too much on nearby data, making boundary sensitive."
    },
    {
      "id": 81,
      "questionText": "Scenario: You combine SVM with bagging ensemble. Benefit?",
      "options": [
        "No change",
        "Overfitting",
        "Higher bias",
        "Reduced variance"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Ensembling multiple SVMs reduces variance and improves generalization."
    },
    {
      "id": 82,
      "questionText": "Scenario: You reduce C drastically. Effect?",
      "options": [
        "Kernel ignored",
        "Perfect accuracy",
        "Wider margin, higher bias",
        "Narrow margin"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Smaller C allows more misclassifications, leading to wider margins."
    },
    {
      "id": 83,
      "questionText": "Scenario: Dataset scaled incorrectly. Decision boundary looks tilted. Reason?",
      "options": [
        "Feature scaling inconsistency",
        "Large γ",
        "Kernel mismatch",
        "Small C"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Inconsistent scaling distorts feature space, altering boundary shape."
    },
    {
      "id": 84,
      "questionText": "Scenario: You use SVM with polynomial kernel on 3D data. Result?",
      "options": [
        "Underfit",
        "Linear separation",
        "Over-generalization",
        "Non-linear surface fit"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Polynomial kernels enable non-linear separation even in 3D."
    },
    {
      "id": 85,
      "questionText": "Scenario: You train SVM on very small dataset. Danger?",
      "options": [
        "High accuracy guaranteed",
        "Underfitting",
        "Overfitting due to few points",
        "Fast convergence"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Small datasets can make SVM overfit due to few support vectors."
    },
    {
      "id": 86,
      "questionText": "Scenario: SVM kernel parameter γ=1e5. What happens?",
      "options": [
        "Underfitting",
        "Extreme overfitting",
        "Stable training",
        "No change"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Very high γ makes the model memorize data, leading to overfitting."
    },
    {
      "id": 87,
      "questionText": "Scenario: SVM hyperplane perfectly separates training points. Danger?",
      "options": [
        "No bias",
        "Perfect generalization",
        "Underfitting",
        "Overfitting likely"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Perfect separation may indicate overfitting unless data is clean."
    },
    {
      "id": 88,
      "questionText": "Scenario: You enable shrinking=True in SVC. Effect?",
      "options": [
        "Faster optimization using heuristics",
        "Slower training",
        "Lower accuracy",
        "No difference"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Shrinking heuristic speeds up convergence during optimization."
    },
    {
      "id": 89,
      "questionText": "Scenario: High dimensional dataset (10000 features). Kernel?",
      "options": [
        "RBF",
        "Linear",
        "Polynomial",
        "Sigmoid"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Linear kernel is efficient in very high-dimensional spaces."
    },
    {
      "id": 90,
      "questionText": "Scenario: Decision boundary too smooth, misclassifying nonlinear data. Fix?",
      "options": [
        "Increase γ or use RBF kernel",
        "Reduce C",
        "Change solver",
        "Add dropout"
      ],
      "correctAnswerIndex": 0,
      "explanation": "Increasing γ adds flexibility to handle complex patterns."
    },
    {
      "id": 91,
      "questionText": "Scenario: You visualize few support vectors only. Meaning?",
      "options": [
        "Training failed",
        "Underfits",
        "Model generalizes well",
        "Overfits"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Fewer support vectors indicate a strong, well-generalized boundary."
    },
    {
      "id": 92,
      "questionText": "Scenario: SVM with RBF kernel used for face recognition. Why suitable?",
      "options": [
        "Less computation",
        "Captures complex non-linear relationships",
        "Linear mapping only",
        "Ignores features"
      ],
      "correctAnswerIndex": 1,
      "explanation": "RBF kernels are effective for non-linear facial feature mapping."
    },
    {
      "id": 93,
      "questionText": "Scenario: Hyperparameter tuning done via grid search. Risk?",
      "options": [
        "Underfitting",
        "Overfitting to validation set",
        "Bias error",
        "Kernel mismatch"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Excessive grid search tuning can overfit to validation data."
    },
    {
      "id": 94,
      "questionText": "Scenario: You observe many support vectors even after tuning. Cause?",
      "options": [
        "Complex data distribution",
        "Low γ",
        "Simpler boundary",
        "Small C"
      ],
      "correctAnswerIndex": 0,
      "explanation": "More support vectors imply complex class boundaries."
    },
    {
      "id": 95,
      "questionText": "Scenario: Kernel trick purpose?",
      "options": [
        "Speed training",
        "Reduce dimensionality",
        "Compute inner products in higher-dimensional space",
        "Add noise"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Kernel trick implicitly computes high-dimensional mappings efficiently."
    },
    {
      "id": 96,
      "questionText": "Scenario: SVM trained with linear kernel on non-linear XOR data. Outcome?",
      "options": [
        "Good generalization",
        "Overfitting",
        "Underfitting",
        "Perfect accuracy"
      ],
      "correctAnswerIndex": 2,
      "explanation": "Linear kernel cannot separate XOR patterns."
    },
    {
      "id": 97,
      "questionText": "Scenario: You set tol=1e-10 in SVM. Effect?",
      "options": [
        "Overfitting",
        "Underfitting",
        "Faster training",
        "Higher precision but slower convergence"
      ],
      "correctAnswerIndex": 3,
      "explanation": "Smaller tolerance increases precision but slows convergence."
    },
    {
      "id": 98,
      "questionText": "Scenario: C=0.01 and γ=1. What’s the likely behavior?",
      "options": [
        "Perfect fit",
        "Underfitting with soft margin",
        "Overfitting",
        "Fast overgeneralization"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Low C allows large margin and misclassifications, causing underfit."
    },
    {
      "id": 99,
      "questionText": "Scenario: RBF kernel used with default params. What’s the effect?",
      "options": [
        "Ignores margin",
        "Fails to train",
        "Depends on feature scaling",
        "Always best choice"
      ],
      "correctAnswerIndex": 2,
      "explanation": "RBF kernel performs well if data is scaled properly."
    },
    {
      "id": 100,
      "questionText": "Scenario: SVM’s decision boundary has maximum margin. Why important?",
      "options": [
        "Increases bias",
        "Improves generalization",
        "Decreases variance",
        "Reduces training speed"
      ],
      "correctAnswerIndex": 1,
      "explanation": "Maximizing margin leads to better generalization and robustness."
    }
  ]
}