WorksheetsMachine Learning CIE
Total questions: 20
Worksheet time: 10mins
A hospital wants to predict whether a patient will be re-admitted within 30 days. The dataset contains: Age, Number of previous visits, Average blood sugar level, Medication adherence score, and Re-admission (Yes/No).
Which algorithm is most suitable for building the first predictive model?
A regression model that assumes a continuous numeric outcome and minimizes Mean Squared Error.
A classification model that estimates the probability of a binary outcome using a sigmoid function.
A dimensionality reduction technique that projects data onto a lower-dimensional subspace.
An unsupervised clustering algorithm that groups patients based on similarity without using target labels.
A retail company has 5 years of data with Date, Product Category, Units Sold, Region, and Season. They want to find hidden purchase patterns such as 'Customers who buy umbrellas often also buy raincoats.' Which method is most suitable?
An algorithm that identifies frequent itemsets and generates association rules such as 'If X, then Y.'
A clustering method that partitions customers into groups based on sales similarity without prior labels.
A predictive model that estimates future sales quantity for each product category using past values.
A technique that reduces correlated sales features into a smaller set of uncorrelated components.
A bank dataset has 99% genuine and 1% fraudulent transactions. Which metric will be least useful for evaluating models?
The overall percentage of correctly predicted transactions, which may look high even if fraud cases are missed.
The proportion of correctly predicted fraudulent transactions out of all actual fraudulent transactions.
The ratio of true positive fraud predictions to the total number of fraud predictions made by the model.
The harmonic mean of precision and recall, balancing both false positives and false negatives.
A researcher collected data on students’ study hours and exam scores. They want to know whether there is a linear relationship between these two variables. Which statistical tool is most appropriate?
A test that checks if the means of two independent groups are significantly different.
A test that calculates the strength and direction of linear association between two continuous variables.
A regression model that predicts categorical outcomes from continuous predictors.
A method that reduces correlated features into a smaller set of uncorrelated components.
A machine learning model shows high performance on training data but very poor results on test data. Which phenomenon is being observed?
The model has underfit due to oversimplification.
The model has high bias and low variance.
The model has overfit by memorizing noise and details in training data.
The model has achieved good generalization.
You have a confusion matrix where: TP=80, FP=20, FN=10, TN=90. Which metric would be highest?
Accuracy, which measures overall proportion of correct predictions.
Precision, which measures how many predicted positives are actually correct.
Recall, which measures how many actual positives are correctly predicted.
F1-Score, which balances precision and recall.
A university wants to study whether gender has an effect on the choice of academic major (Science, Commerce, Arts). Which test is suitable?
A regression model assuming continuous outcomes.
A correlation test for continuous variables.
A paired t-test for dependent samples.
A chi-square test for independence between categorical variables.
A company collects data on employee experience (in years) and their salary (in ₹). The HR team wants to measure whether employees with more years of experience tend to earn higher salaries. They also want the result to be independent of the units of measurement, so it can be compared with other company datasets.
Use a measure that shows direction and degree of linear association after standardizing the variables.
Use a measure that gives raw joint variability in years and rupees, which depends on units of data.
Use a test that checks difference in means of two groups.
Use a method that reduces correlated features into fewer components.
In a school, researchers are studying students’ height (cm) and weight (kg). They only want to know whether taller students generally tend to weigh more, but they don’t care about the actual numerical scale.
Use a metric that gives standardized relationship, bounded between -1 and +1.
Use a metric that multiplies deviations in original units, producing values that are not standardized.
Use a test that compares height averages of two groups of students.
Use a method that predicts one variable using multiple independent predictors.
If correlation between X and Y is exactly 0, what does it imply?
X and Y are independent.
X and Y must always vary in opposite directions.
X and Y have zero covariance but strong dependence.
X and Y have no linear relationship, but they may still have a non-linear relationship.
Two datasets are analyzed:
Dataset A: Height (cm) vs Weight (kg) → reported value = 0.85
Dataset B: Monthly Salary (₹) vs Expenditure (₹) → reported value = 20,000
Which interpretation is the most appropriate?
The second dataset must reflect a stronger relationship since its numeric value is much larger than the first dataset’s reported value.
Both datasets clearly represent equally strong associations because both numbers are relatively high in magnitude, regardless of scale differences.
Dataset A shows a strong linear relationship, while Dataset B’s value cannot be directly compared because covariance depends on units.
The value reported for the first dataset can be directly treated as the same kind of measure as the second dataset if expressed in the right units.
Why is correlation preferred over covariance when comparing relationships between different pairs of variables?
Because it generates values that reflect co-movement only in the raw measurement scales, so the interpretation directly depends on the magnitude of the original data.
Because it is based on a bounded metric, the interpretation remains consistent across datasets, unlike raw variability which cannot be directly compared.
Because it avoids dependence on the absolute variance of each variable, thereby ensuring that two relationships measured in different scales can be meaningfully contrasted.
Because correlation removes the units of measurement by normalizing with standard deviations, making relationships comparable across datasets.
A real estate analyst collects house prices for a city. They notice a few extremely expensive mansions in the dataset and want a measure of variability that is not affected by these outliers to summarize the spread of house prices.
Use the standard deviation to calculate average deviation from the mean, which will heavily reflect extreme prices.
Use the range (maximum – minimum) to get a sense of spread including all values.
Use a measure that calculates the IQR to understand the central spread while ignoring extreme values.
Use the mean as a measure of central tendency to summarize the typical house price.
A data scientist builds a linear regression model to predict monthly sales based on advertising spend and store footfall. After fitting the model, they want a metric that captures average squared prediction error and a measure of how well predictors explain variability in sales, accounting for the number of predictors.
Use median absolute error and ignore how many predictors are in the model.
Only check the correlation coefficient between predicted and actual sales without considering prediction errors.
Use only mean squared error without considering how many variables are included, as it automatically adjusts for model complexity.
Calculate MSE for sales, and also check a goodness-of-fit measure that adjusts for the number of independent variables in the model.
A teacher collects marks of 100 students. The dataset has a few extremely high scores. She wants a measure of central tendency that best represents the typical student without being skewed by these extreme scores, and another measure that tells the most frequently obtained score.
Use only mean, as it automatically handles extreme scores and identifies common scores.
Use mode for average performance and range for frequency.
Use median to represent the typical performance and mode to find the most common score.
Use mean to represent typical performance and median for the most common score.
A model predicts monthly sales (in units) for 4 stores. Actual vs predicted:
What is the Mean Squared Error (MSE)?
5
4
6
8
A dataset of 5 students’ study hours and test scores:
No correlation, approximately 0
Perfect correlation, exactly 1
Negative correlation, approximately -0.97
Positive correlation, approximately 0.97
In a study, 100 people are surveyed for exposure to a chemical and disease occurrence:
What is the odds ratio for disease in exposed vs non-exposed?
6
3
2
0.5
A bank is building a model to detect fraudulent credit card transactions. Out of 10,000 transactions, only 50 are actually fraudulent. A false positive (flagging a genuine transaction as fraud) is annoying for customers, but a false negative (missing a fraud) could cause a large financial loss.
Which evaluation metric should the bank prioritize while tuning the model?
Precision, because it measures the proportion of predicted fraud cases that are actually fraudulent.
Recall, because it measures the proportion of actual fraud cases correctly identified, reducing financial losses.
Accuracy, because it measures overall proportion of correct predictions, regardless of class imbalance.
F1-Score, because it averages precision and recall without considering business impact.
A company wants to build a spam email filter. If a legitimate email is marked as spam, employees may miss important messages (annoying, but recoverable). If a spam email is not detected, it may contain phishing or malware.
Which evaluation metric should the company focus on?
Accuracy, because it measures overall proportion of correctly classified emails, including both spam and non-spam.
Specificity, because it only measures non-spam classification correctness.
Recall, because it measures proportion of actual spam emails correctly detected, protecting employees from phishing.
Precision, because it measures proportion of emails flagged as spam that are truly spam, reducing false alarms.
