1.

Install and Set Up Python and Essential Libraries

(NumPy, pandas)
Q1: Why is Python a preferred language for data science and machine
learning?

Answer:
Python is preferred because:
✔ Easy to Learn: Simple syntax, readable code.
✔ Rich Ecosystem: Libraries like NumPy, pandas, scikit-learn, TensorFlow.
✔ Community Support: Large open-source community.
✔ Versatility: Used in web development, automation, and AI.
✔ Integration: Works well with databases, big data tools (Spark), and visualization
libraries (Matplotlib, Seaborn).

Q2: What is the purpose of pip in Python, and how do you install a library
using it?

Answer:

 pip is Python’s package installer.

 To install a library:

pip install numpy pandas

 To upgrade:
pip install --upgrade pandas

Q3: Explain the difference between NumPy and pandas.

Answer:
NumPy pandas

Used for numerical computations (arrays,

Used for structured data (DataFrames, Series).
matrices).

Optimized for math operations (linear Optimized for data manipulation (filtering,
algebra). grouping).

[Link]() and [Link]() are core

[Link]() is the core data structure.
structures.

Q4: How do you check the installed version of a Python library like
pandas?

Answer:

import pandas as pd
print(pd.__version__)

or via terminal:

pip show pandas

Q5: What is a virtual environment, and why is it useful?

Answer:

 A virtual environment isolates Python dependencies for different projects.

 Why use it?
✔ Avoids version conflicts between projects.
✔ Ensures reproducibility.
 How to create one?
python -m venv myenv
source myenv/bin/activate # Linux/Mac
myenv\Scripts\activate # Windows
2. Introduction to Scikit-Learn
Q1: What is scikit-learn, and what are its key features?

Answer:

 Scikit-learn is a Python ML library for:

✔ Classification (k-NN, SVM, Decision Trees).
✔ Regression (Linear, Ridge, Lasso).
✔ Clustering (K-Means, DBSCAN).
✔ Preprocessing (Scaling, Encoding).
 Features:
✔ Open-source, easy-to-use API.
✔ Built on NumPy, SciPy, and Matplotlib.

Q2: Name some common ML algorithms in scikit-learn.

Answer:

 Supervised Learning:

o LinearRegression, LogisticRegression

o DecisionTreeClassifier, RandomForest

o SVM, k-NN

 Unsupervised Learning:

o KMeans, DBSCAN

o PCA (Dimensionality Reduction)

Q3: How does scikit-learn differ from TensorFlow/PyTorch?

Answer:

Scikit-Learn TensorFlow/PyTorch

Traditional ML (small datasets). Deep Learning (neural networks).

Scikit-Learn TensorFlow/PyTorch

Simpler API. Complex, GPU-accelerated.

No GPU support. Optimized for GPU/TPU.

Q4: What are the main modules in scikit-learn?

Answer:

 sklearn.model_selection (train-test split, cross-validation).

 [Link] (scaling, encoding).

 [Link] (accuracy, MSE, confusion matrix).

 sklearn.linear_model (LinearRegression, LogisticRegression).

Q5: Can scikit-learn be used for deep learning?

Answer:
❌ No, scikit-learn is for traditional ML. For deep learning, use:
✔ TensorFlow / Keras
✔ PyTorch

3. Install and Set Up Scikit-Learn

Q1: How to install scikit-learn?

Answer:

pip install scikit-learn

or with conda:

conda install scikit-learn

Q2: What are the dependencies of scikit-learn?

Answer:

 NumPy
 SciPy
 Matplotlib (for visualization)
 Joblib (for model saving)

Q3: How to verify scikit-learn installation?

Answer:

import sklearn
print(sklearn.__version__)

Q4: What is Jupyter Notebook, and why use it?

Answer:

 Interactive Python environment for data analysis.

 Why use it?
✔ Combines code, visualizations, and text.
✔ Great for exploratory data analysis (EDA).

Q5: Alternative tools alongside scikit-learn?

Answer:

 XGBoost (Gradient Boosting).

 StatsModels (Statistical modeling).
 LightGBM (Optimized decision trees).

4. Load and Explore CSV/Excel Data with Pandas

Q1: How to read a CSV file in pandas?
Answer:

import pandas as pd
data = pd.read_csv("[Link]")

Q2: Common pandas functions for exploration?

Answer:

 [Link]() → First 5 rows.

 [Link]() → Statistics (mean, std, min, max).

 [Link]() → Data types and missing values.

Q3: How to check for missing values?

Answer:

[Link]().sum()

Q4: Difference between loc and iloc?

Answer:

 loc → Label-based indexing ([Link][0, 'column']).

 iloc → Integer-based indexing ([Link][0, 1]).

Q5: How to filter rows in a DataFrame?

Answer:

filtered_data = data[data['column'] > 50]

5. Data Visualization with Matplotlib/Seaborn

Q1: Difference between Matplotlib and Seaborn?
Answer:

 Matplotlib: Basic, customizable plots.

 Seaborn: Built on Matplotlib, prettier statistical plots.

Q2: How to create a scatter plot?

Answer:

import [Link] as plt

[Link](x=data['X'], y=data['Y'])
[Link]()

Q3: Plots for distributions?

Answer:

 Histogram ([Link]()).
 Boxplot ([Link]()).

Q4: How to customize plot labels?

Answer:

[Link]("X-axis")
[Link]("Y-axis")
[Link]("Scatter Plot")

Q5: Bar Chart vs. Pie Chart?

Answer:

 Bar Chart: Compare categories (e.g., sales per month).

 Pie Chart: Show proportions (e.g., market share %).

(Continued in next message for remaining answers...)

Would you like me to proceed with the rest? 🚀

yes

Answers to Viva Questions (Continued)

6. Handling Missing Data, Encoding, and Feature Scaling

Q1: Strategies to handle missing data?

Answer:
✔ Deletion: dropna() (if few missing values).
✔ Imputation:

 Mean/Median (SimpleImputer).
 Mode (for categorical data).
 Advanced: KNN imputation.

Q2: What is one-hot encoding?

Answer:

 Converts categorical variables into binary columns.

 Example:
pd.get_dummies(data['category'])

 When to use? When categories are not ordinal (e.g., colors: Red, Green, Blue).

Q3: Why is feature scaling important?

Answer:

 Ensures all features contribute equally to distance-based algorithms (k-NN, SVM).

 Prevents dominance of high-magnitude features.
Q4: StandardScaler vs. MinMaxScaler?

Answer:

StandardScaler MinMaxScaler

Scales to mean=0, std=1. Scales to range [0, 1].

Good for Gaussian data. Good for bounded data.

Q5: How to encode categorical variables?

Answer:

 Label Encoding (Ordinal):

from [Link] import LabelEncoder
le = LabelEncoder()
data['category'] = le.fit_transform(data['category'])

 One-Hot Encoding (Nominal):

pd.get_dummies(data)

7. k-Nearest Neighbors (k-NN) Classifier

Q1: Basic principle of k-NN?

Answer:

 Classifies a data point based on majority vote of its k nearest neighbors.

 Steps:

1. Compute distances (Euclidean/Manhattan).

2. Find k nearest points.
3. Assign majority class.

Q2: How does k affect performance?

Answer:

 Small k:
✔ High variance (overfitting).
✔ Sensitive to noise.
 Large k:
✔ High bias (underfitting).
✔ Smoother decision boundaries.

Q3: Distance metrics in k-NN?

Answer:

 Euclidean: Default (sqrt(∑(x_i - y_i)²)).

 Manhattan: Sum of absolute differences (∑|x_i - y_i|).
 Minkowski: Generalization of both.

Q4: Train-test split in scikit-learn?

Answer:

from sklearn.model_selection import train_test_split

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)

Q5: Evaluation metrics for classification?

Answer:

 Accuracy: (TP + TN) / Total.

 Precision: TP / (TP + FP).
 Recall: TP / (TP + FN).
 F1-Score: Harmonic mean of precision/recall.
8. Linear Regression Model
Q1: Mathematical formulation?

Answer:
y = β₀ + β₁x₁ + β₂x₂ + ... + βₙ xₙ + ε

 y: Target variable.

 β₀: Intercept.

 β₁...βₙ : Coefficients.

 ε: Error term.

Q2: How to interpret coefficients?

Answer:

 β₁: For 1-unit increase in x₁, y changes by β₁ (holding other features constant).
 Example: If β₁ = 2.5, y increases by 2.5 for every 1-unit increase in x₁.

Q3: fit() vs. predict()?

Answer:

 fit(): Trains the model on data ([Link](X_train, y_train)).

 predict(): Generates predictions (y_pred = [Link](X_test)).

Q4: What is Mean Squared Error (MSE)?

Answer:

 Average squared difference between predicted and actual values.

 Formula:
from [Link] import mean_squared_error
mse = mean_squared_error(y_test, y_pred)

Q5: How to check for overfitting?

Answer:

 High training accuracy, low test accuracy → Overfitting.

 Solutions:
✔ Regularization (Ridge/Lasso).
✔ Cross-validation.

9. Decision Tree Classifier

Q1: How does a decision tree make splits?

Answer:

 Splits based on:

o Gini Impurity: Measures misclassification probability.

o Entropy: Measures disorder (lower = better split).

Q2: What is Gini impurity?

Answer:

 Probability of incorrect classification for a randomly chosen element.

 Formula: Gini = 1 - ∑(p_i)² (where p_i is class proportion).

Q3: How to visualize a decision tree?

Answer:

from [Link] import plot_tree

plot_tree(model, feature_names=[Link], filled=True)

Q4: Pros and cons of decision trees?

Answer:
Pros Cons

Easy to interpret. Prone to overfitting.

No need for scaling. Unstable (small changes → different tree).

Q5: What is pruning?

Answer:

 Reduces tree size by removing unnecessary splits.

 Methods:

o Pre-pruning (limit depth/min_samples_split).

o Post-pruning (cost-complexity pruning).

10. K-Means Clustering

Q1: Objective of K-Means?

Answer:

 Partition data into k clusters where each point belongs to the nearest centroid.
 Goal: Minimize within-cluster variance.

Q2: How to choose optimal k?

Answer:

 Elbow Method: Plot Inertia (sum of squared distances) vs. k.

 Silhouette Score: Measures cluster separation.

Q3: Role of centroid?

Answer:
  Mean position of all points in a cluster.
 Updated iteratively to minimize distances.

Q4: Initialization sensitivity?

Answer:

 Random centroids can lead to suboptimal clusters.

 Solution: Use k-means++ (smart initialization).

Q5: Limitations of K-Means?

Answer:
❌ Assumes spherical clusters.
❌ Sensitive to outliers.
❌ Requires predefined k.

Summary Table: Key Scikit-Learn Methods

Task Function

Train-test split train_test_split()

Feature scaling StandardScaler(), MinMaxScaler()

k-NN Classifier KNeighborsClassifier()

Linear Regression LinearRegression()

Decision Tree DecisionTreeClassifier()

K-Means KMeans()