Classical Prediction Models with Scikit-learn MCQ Questions with Answers (Latest 2026)
Practice Classical Prediction Models with Scikit-learn MCQ questions with detailed explanations and clear answer validation. These MCQs help you revise core concepts, compare close options, and improve accuracy for interviews, certification exams, and technical screening rounds. Use this updated 2026 set to strengthen fundamentals and confidence.
Q1. What type of task predicts a continuous numeric value?
Select an answer to check.
Answer: Regression
Here, Regression is the right choice. Regression models estimate continuous targets such as price or demand. It aligns directly with what the question asks about what type of task predicts a continuous numeric. A quick elimination of partially true options helps confirm it.
Q2. What type of task predicts a category label?
Select an answer to check.
Answer: Classification
In this case, Classification is correct. Classification assigns inputs to predefined classes. It aligns directly with what the question asks about what type of task predicts a category label. A quick elimination of partially true options helps confirm it.
Q3. In scikit-learn, what does fit() do?
Select an answer to check.
Answer: Trains the model on data
The best option here is Trains the model on data. Fit() learns parameters from training features and targets. It aligns directly with what the question asks about in scikit-learn, what does fit() do. A quick elimination of partially true options helps confirm it.
Q4. In scikit-learn, what does predict() do?
Select an answer to check.
Answer: Returns model outputs for new inputs
For this question, Returns model outputs for new inputs is correct. Predict() uses learned parameters to infer target values or classes. It aligns directly with what the question asks about in scikit-learn, what does predict() do. A quick elimination of partially true options helps confirm it.
Q5. Which utility is commonly used to split data into train and test sets?
Select an answer to check.
Answer: train_test_split
train_test_split is the correct answer here. Train_test_split creates holdout subsets for evaluation. It aligns directly with what the question asks about which utility is commonly used to split data. A quick elimination of partially true options helps confirm it.
Q6. Why set random_state in experiments?
Select an answer to check.
Answer: For reproducibility
Here, For reproducibility is the right choice. Random_state makes random operations deterministic across runs. This matches the core idea being tested around why set random_state in experiments. A quick elimination of partially true options helps confirm it.
Q7. Which scaler standardizes features to zero mean and unit variance?
Select an answer to check.
Answer: StandardScaler
In this case, StandardScaler is correct. StandardScaler transforms each feature to mean 0 and variance 1. This matches the core idea being tested around which scaler standardizes features to zero mean and. A quick elimination of partially true options helps confirm it.
Q8. Which scaler maps values to a fixed range like [0,1]?
Select an answer to check.
Answer: MinMaxScaler
The best option here is MinMaxScaler. MinMaxScaler rescales features based on observed min and max. This matches the core idea being tested around which scaler maps values to a fixed range. A quick elimination of partially true options helps confirm it.
Q9. How should nominal categorical features usually be encoded?
Select an answer to check.
Answer: One-hot encoding
For this question, One-hot encoding is correct. One-hot encoding avoids implying false order among categories. This matches the core idea being tested around how should nominal categorical features usually be encoded. A quick elimination of partially true options helps confirm it.
Q10. Why is LabelEncoder usually not ideal for input categorical columns?
Select an answer to check.
Answer: It imposes ordinal meaning
It imposes ordinal meaning is the correct answer here. Integer labels can suggest artificial distance/order for categories. This matches the core idea being tested around why is labelencoder usually not ideal for input. A quick elimination of partially true options helps confirm it.
Q11. What is the main benefit of Pipeline in scikit-learn?
Select an answer to check.
Answer: Combines preprocessing and model steps safely
Here, Combines preprocessing and model steps safely is the right choice. Pipelines make workflows cleaner and help prevent data leakage. That is exactly the concept behind what is the main benefit of pipeline in in this context. A quick elimination of partially true options helps confirm it.
Q12. Which tool applies different preprocessing to different column groups?
Select an answer to check.
Answer: ColumnTransformer
In this case, ColumnTransformer is correct. ColumnTransformer routes specified columns through separate transforms. That is exactly the concept behind which tool applies different preprocessing to different column in this context. A quick elimination of partially true options helps confirm it.
Q13. LinearRegression minimizes which quantity by default?
Select an answer to check.
Answer: Residual sum of squares
The best option here is Residual sum of squares. Ordinary least squares fits coefficients minimizing squared residuals. That is exactly the concept behind linearregression minimizes which quantity by default in this context. A quick elimination of partially true options helps confirm it.
Q14. Ridge regression adds what type of penalty?
Select an answer to check.
Answer: L2 penalty
For this question, L2 penalty is correct. Ridge uses L2 regularization to shrink coefficients. That is exactly the concept behind ridge regression adds what type of penalty in this context. A quick elimination of partially true options helps confirm it.
Q15. Lasso regression adds what type of penalty?
Select an answer to check.
Answer: L1 penalty
L1 penalty is the correct answer here. Lasso can drive some coefficients exactly to zero for sparsity. That is exactly the concept behind lasso regression adds what type of penalty in this context. A quick elimination of partially true options helps confirm it.
Q16. ElasticNet combines which penalties?
Select an answer to check.
Answer: L1 and L2
Here, L1 and L2 is the right choice. ElasticNet balances variable selection and shrinkage. It fits the requirement in the prompt about elasticnet combines which penalties. A quick elimination of partially true options helps confirm it.
Q17. LogisticRegression in scikit-learn is mainly used for:
Select an answer to check.
Answer: Classification
In this case, Classification is correct. Despite its name, logistic regression is a classification algorithm. It fits the requirement in the prompt about logisticregression in scikit-learn is mainly used for:. A quick elimination of partially true options helps confirm it.
Q18. Binary logistic regression models class probability using:
Select an answer to check.
Answer: Sigmoid function
The best option here is Sigmoid function. The sigmoid maps linear scores to probabilities in [0,1]. It fits the requirement in the prompt about binary logistic regression models class probability using:. A quick elimination of partially true options helps confirm it.
Q19. For binary classification, a common default probability threshold is:
Select an answer to check.
Answer: 0.5
For this question, 0.5 is correct. Many workflows use 0.5 by default, though threshold tuning is common. It fits the requirement in the prompt about for binary classification, a common default probability threshold. A quick elimination of partially true options helps confirm it.
Q20. For multiclass logistic regression, which strategy can be used?
Select an answer to check.
Answer: Multinomial
Multinomial is the correct answer here. Multinomial logistic handles multiple classes directly. It fits the requirement in the prompt about for multiclass logistic regression, which strategy can be. A quick elimination of partially true options helps confirm it.
Q21. In KNeighborsClassifier, which hyperparameter controls neighborhood size?
Select an answer to check.
Answer: n_neighbors
Here, n_neighbors is the right choice. N_neighbors defines how many nearby points vote. This is the most accurate statement for in kneighborsclassifier, which hyperparameter controls neighborhood size. A quick elimination of partially true options helps confirm it.
Q22. Increasing k in KNN often causes:
Select an answer to check.
Answer: Higher bias, lower variance
In this case, Higher bias, lower variance is correct. Larger neighborhoods smooth decision boundaries. This is the most accurate statement for increasing k in knn often causes:. A quick elimination of partially true options helps confirm it.
Q23. Which hyperparameter limits tree growth depth?
Select an answer to check.
Answer: max_depth
The best option here is max_depth. Max_depth constrains complexity and can reduce overfitting. This is the most accurate statement for which hyperparameter limits tree growth depth. A quick elimination of partially true options helps confirm it.
Q24. A common split criterion in decision trees for classification is:
Select an answer to check.
Answer: Gini impurity
For this question, Gini impurity is correct. Trees choose splits that reduce impurity like Gini or entropy. This is the most accurate statement for a common split criterion in decision trees for. A quick elimination of partially true options helps confirm it.
Q25. RandomForest is primarily based on:
Select an answer to check.
Answer: Bagging many trees
Bagging many trees is the correct answer here. Random forests average many decorrelated trees for robustness. This is the most accurate statement for randomforest is primarily based on:. A quick elimination of partially true options helps confirm it.
Q26. ExtraTrees increases randomness by:
Select an answer to check.
Answer: Choosing random split thresholds
Here, Choosing random split thresholds is the right choice. ExtraTrees samples splits more randomly than standard random forest. It aligns directly with what the question asks about extratrees increases randomness by:. The other options are either incomplete or contextually incorrect.
Q27. GradientBoosting builds trees:
Select an answer to check.
Answer: Sequentially to correct errors
In this case, Sequentially to correct errors is correct. Boosting adds weak learners that fit residual errors. It aligns directly with what the question asks about gradientboosting builds trees:. The other options are either incomplete or contextually incorrect.
Q28. AdaBoost focuses learning by:
Select an answer to check.
Answer: Reweighting misclassified samples
The best option here is Reweighting misclassified samples. AdaBoost increases focus on previously misclassified points. It aligns directly with what the question asks about adaboost focuses learning by:. The other options are either incomplete or contextually incorrect.
Q29. Which statement about XGBoost and scikit-learn is correct?
Select an answer to check.
Answer: XGBoost is an external library with sklearn-style API
For this question, XGBoost is an external library with sklearn-style API is correct. XGBoost is separate but interoperates with sklearn patterns. It aligns directly with what the question asks about which statement about xgboost and scikit-learn is correct. The other options are either incomplete or contextually incorrect.
Q30. In SVM, larger C generally means:
Select an answer to check.
Answer: Weaker regularization and stricter fitting
Weaker regularization and stricter fitting is the correct answer here. High C penalizes misclassification more, fitting training data more closely. It aligns directly with what the question asks about in svm, larger c generally means:. The other options are either incomplete or contextually incorrect.
Q31. Kernel trick in SVM enables:
Select an answer to check.
Answer: Nonlinear decision boundaries in transformed space
Here, Nonlinear decision boundaries in transformed space is the right choice. Kernels compute similarities without explicit high-dimensional mapping. This matches the core idea being tested around kernel trick in svm enables:. The other options are either incomplete or contextually incorrect.
Q32. In RBF SVM, gamma controls:
Select an answer to check.
Answer: Influence radius of training points
In this case, Influence radius of training points is correct. Higher gamma means narrower influence and potentially more complex boundaries. This matches the core idea being tested around in rbf svm, gamma controls:. The other options are either incomplete or contextually incorrect.
Q33. Naive Bayes assumes features are:
Select an answer to check.
Answer: Conditionally independent given class
The best option here is Conditionally independent given class. This simplifying assumption enables fast probabilistic classification. This matches the core idea being tested around naive bayes assumes features are:. The other options are either incomplete or contextually incorrect.
Q34. GaussianNB is appropriate when features are roughly:
Select an answer to check.
Answer: Normally distributed continuous values
For this question, Normally distributed continuous values is correct. GaussianNB models each feature likelihood with Gaussian distributions. This matches the core idea being tested around gaussiannb is appropriate when features are roughly:. The other options are either incomplete or contextually incorrect.
Q35. MultinomialNB is commonly used for:
Select an answer to check.
Answer: Count-based text features
Count-based text features is the correct answer here. It works well with term-frequency style nonnegative counts. This matches the core idea being tested around multinomialnb is commonly used for:. The other options are either incomplete or contextually incorrect.
Q36. BernoulliNB is designed for:
Select an answer to check.
Answer: Binary/boolean features
Here, Binary/boolean features is the right choice. BernoulliNB models feature presence/absence. That is exactly the concept behind bernoullinb is designed for: in this context. The other options are either incomplete or contextually incorrect.
Q37. What does a confusion matrix summarize?
Select an answer to check.
Answer: Prediction vs actual class counts
In this case, Prediction vs actual class counts is correct. It shows true/false positives and negatives by class. That is exactly the concept behind what does a confusion matrix summarize in this context. The other options are either incomplete or contextually incorrect.
Q38. Why can accuracy be misleading on imbalanced data?
Select an answer to check.
Answer: Majority class can dominate score
The best option here is Majority class can dominate score. High accuracy can hide poor minority-class detection. That is exactly the concept behind why can accuracy be misleading on imbalanced data in this context. The other options are either incomplete or contextually incorrect.
Q39. Precision is defined as:
Select an answer to check.
Answer: TP/(TP+FP)
For this question, TP/(TP+FP) is correct. Precision measures correctness among predicted positives. That is exactly the concept behind precision is defined as: in this context. The other options are either incomplete or contextually incorrect.
Q40. Recall is defined as:
Select an answer to check.
Answer: TP/(TP+FN)
TP/(TP+FN) is the correct answer here. Recall measures how many actual positives were found. That is exactly the concept behind recall is defined as: in this context. The other options are either incomplete or contextually incorrect.
Q41. F1-score is the:
Select an answer to check.
Answer: Harmonic mean of precision and recall
Here, Harmonic mean of precision and recall is the right choice. F1 balances precision and recall in one metric. It fits the requirement in the prompt about f1-score is the:. The other options are either incomplete or contextually incorrect.
Q42. ROC-AUC mainly evaluates:
Select an answer to check.
Answer: Ranking quality across thresholds
In this case, Ranking quality across thresholds is correct. AUC summarizes true-positive vs false-positive tradeoff. It fits the requirement in the prompt about roc-auc mainly evaluates:. The other options are either incomplete or contextually incorrect.
Q43. For heavy class imbalance, which curve is often more informative?
Select an answer to check.
Answer: Precision-Recall curve
The best option here is Precision-Recall curve. PR emphasizes positive class performance under imbalance. It fits the requirement in the prompt about for heavy class imbalance, which curve is often. The other options are either incomplete or contextually incorrect.
Q44. cross_val_score is used to:
Select an answer to check.
Answer: Evaluate model across folds
For this question, Evaluate model across folds is correct. It returns validation scores from repeated train/validation splits. It fits the requirement in the prompt about cross_val_score is used to:. The other options are either incomplete or contextually incorrect.
Q45. KFold cross-validation splits data into:
Select an answer to check.
Answer: k train-test partitions
k train-test partitions is the correct answer here. Each fold acts once as validation while others train. It fits the requirement in the prompt about kfold cross-validation splits data into:. The other options are either incomplete or contextually incorrect.
Q46. StratifiedKFold is useful because it:
Select an answer to check.
Answer: Preserves class proportions in each fold
Here, Preserves class proportions in each fold is the right choice. It keeps target distribution stable in classification folds. This is the most accurate statement for stratifiedkfold is useful because it:. The other options are either incomplete or contextually incorrect.
Q47. GridSearchCV performs:
Select an answer to check.
Answer: Exhaustive search over given param grid
In this case, Exhaustive search over given param grid is correct. It evaluates all specified hyperparameter combinations. This is the most accurate statement for gridsearchcv performs:. The other options are either incomplete or contextually incorrect.
Q48. RandomizedSearchCV differs by:
Select an answer to check.
Answer: Sampling limited random combinations
The best option here is Sampling limited random combinations. It can find good settings faster in large search spaces. This is the most accurate statement for randomizedsearchcv differs by:. The other options are either incomplete or contextually incorrect.
Q49. Setting n_jobs=-1 generally means:
Select an answer to check.
Answer: Use all available CPU cores
For this question, Use all available CPU cores is correct. Many sklearn estimators/searches parallelize with n_jobs. This is the most accurate statement for setting n_jobs=-1 generally means:. The other options are either incomplete or contextually incorrect.
Q50. The scoring parameter in model selection controls:
Select an answer to check.
Answer: Optimization metric for selecting best model
Optimization metric for selecting best model is the correct answer here. Best hyperparameters are chosen based on scoring metric. This is the most accurate statement for the scoring parameter in model selection controls:. The other options are either incomplete or contextually incorrect.