1️⃣ Question 1

SVM with binary decisions extended to 3 classes:

• ❌ Combining supervised + unsupervised

• ❌ One classifier per class (one-vs-rest)

• ❌ Single classifier

• ✅ One classifier per pair of classes (one-vs-one)

Explanation:
SVM commonly uses one-vs-one for multi-class classification.

---

2️⃣ Question 2

Why use median instead of mean for skewed salary data?

• ❌ Minimizes MSE

• ✅ Reduces impact of extreme values

• ❌ Mean is inaccurate

• ❌ Mean is hard to compute

---

3️⃣ Question 3

Decision tree increases complexity → What happens?

• ✅ Bias decreases, variance increases

• ❌ Bias increases, variance decreases

• ❌ Both constant

• ❌ Both decrease

---

4️⃣ Question 4

Best ML task for detecting unusual banking transactions:

• ❌ Predicting monthly trends

• ✅ Identifying patterns that deviate from normal transactions (anomaly detection)

• ❌ Predefined risk classification

• ❌ Grouping transactions

---

5️⃣ Question 5

Productivity increases, slows, levels off:

• ❌ Exponential

• ❌ Logarithmic

• ✅ Polynomial regression

• ❌ Linear

Explanation:
Nonlinear trend that increases then stabilizes → polynomial regression.

---

6️⃣ Question 6

Binary classification based on proximity to neighbors:

• ❌ Logistic regression

• ❌ Decision tree

• ✅ K-nearest neighbors (KNN)

• ❌ SVM

---

7️⃣ Question 7

Benefit of PCA before clustering:

• ✅ Transforms features into principal axes with highest variance

• ❌ Removes all less important features

• ❌ Automatically segments customers

• ❌ Reduces to one component

---

8️⃣ Question 8

Fast, scalable logistic regression training method:

• ❌ Backpropagation

• ❌ Grid search

• ❌ Least squares

• ✅ Stochastic Gradient Descent (SGD)

---

9️⃣ Question 9

Model misclassifies loyal customers as churn risks:

• ❌ PCA

• ❌ SVM

• ❌ More churn data

• ✅ Adjust the classification threshold

---

🔟 Question 10

Clustering method starting with individuals → merging:

• ❌ Density-based

• ❌ Divisive

• ✅ Agglomerative clustering

• ❌ Partition-based

---

1️⃣1️⃣ Question 11

Why is DBSCAN ideal for the marketing team’s use case?

• ❌ Daily travel routines

• ❌ Purchase trends forecasting

• ❌ Detect satellite green cover

• ✅ Isolate rare sensor events in IoT data

Explanation:
DBSCAN is excellent for outlier detection, not forecasting or segmentation.

---

1️⃣2️⃣ Question 12

Method preserving local + global structure:

• ❌ PCA

• ✅ UMAP

• ❌ Not used

• ❌ t-SNE

Explanation:
t-SNE keeps local structure only;
UMAP preserves both local AND global.

---

1️⃣3️⃣ Question 13

Tool for interactive ML dashboards:

• ❌ Pandas

• ❌ Matplotlib

• ❌ Scikit-learn

• ❌ NumPy

Correct Intended Answer:
➡️ This question is tricky.
Dashboards = Matplotlib isn’t interactive, Pandas & NumPy not dashboards, scikit-learn isn’t visualization.

Correct tool (from course context):
✅ Matplotlib (used for ML visualization)

---

1️⃣4️⃣ Question 14

Difference between ML & traditional programming:

• ❌ Writes code faster

• ❌ Generates random rules

• ✅ Learns from data to make predictions

• ❌ Uses hand-coded rules

---

1️⃣5️⃣ Question 15

Matrix operations and linear algebra:

• ❌ Scikit-learn

• ❌ Pandas

• ❌ Matplotlib

• ✅ NumPy

---

🧾 Summary Table

---