ML- Oral Questions and Answers

1. What is Machine Learning (ML)?
Machine Learning is a branch of AI that enables computers to learn patterns from data and
make predictions or decisions without being explicitly programmed for every task.
2. Why do we need Machine Learning?
ML is useful when problems involve large amounts of data, complex patterns, and predictions
that are difficult to solve using fixed rules.
3. What is the difference between traditional programming and ML?
Traditional programming uses data and explicitly written rules to produce outputs. ML uses
training data to learn a model that produces predictions.
4. What is the difference between AI and ML?
AI is the broader field of creating intelligent systems, whereas ML is a subset of AI that
learns patterns from data.
5. Give three real-life applications of ML.
Email spam detection, product recommendations, and disease prediction.
6. What are the main types of Machine Learning?
Supervised learning, unsupervised learning, reinforcement learning, and semi-supervised
learning.
7. What is supervised learning?
It learns from labeled data, where the correct output is provided during training.
8. What is unsupervised learning?
It discovers patterns or groups in unlabeled data.
9. What is reinforcement learning?
An agent learns to make decisions by interacting with an environment and receiving rewards
or penalties.
10. What is semi-supervised learning?
It uses a small amount of labeled data along with a larger amount of unlabeled data.
11. What is the difference between classification and regression?
Classification predicts categories, whereas regression predicts numerical values.
12. What is an ML workflow?
It includes problem definition, data collection, preprocessing, feature engineering, data
splitting, model training, evaluation, and deployment.
13. What is structured data?
Data organized in a predefined format, usually rows and columns, such as a student marks
spreadsheet.

14. What is unstructured data?
Data without a fixed tabular structure, such as images, audio recordings, and videos.
15. What is semi-structured data?
Data that has organizational tags or keys but does not follow a rigid table structure, such as
JSON or XML.
16. What are common methods of data collection?
Surveys, sensors, experiments, databases, APIs, and web data collection.
17. What is data integration?
Combining data from multiple sources into a unified dataset.
18. What is data preprocessing?
Preparing raw data for ML by handling missing values, duplicates, inconsistent formats, and
other data quality issues.
19. What is feature scaling?
It transforms numerical features to comparable scales using techniques such as
standardization and min-max normalization.
20. What is the difference between normalization and standardization?
Normalization often scales values to a fixed range, such as 0 to 1. Standardization transforms
values to have approximately zero mean and unit standard deviation.
21. Why do we split data into training and testing sets?
Training data is used to fit the model, while testing data evaluates its performance on unseen
examples.
22. What is overfitting?
Overfitting occurs when a model learns training data too closely, including noise, and
performs poorly on unseen data.
23. What is underfitting?
Underfitting occurs when a model is too simple to capture the important patterns in the data.
24. How can overfitting be reduced?
By using regularization, cross-validation, more representative training data, pruning, or
simpler models.
25. What is dimensionality reduction?
It reduces the number of features while retaining as much relevant information as possible.
26. What is PCA?
Principal Component Analysis transforms features into new uncorrelated components that
capture maximum variance.
27. What is LDA?
Linear Discriminant Analysis finds projections that help separate known classes. It is a
supervised technique commonly used for dimensionality reduction.

28. What is t-SNE?
t-distributed Stochastic Neighbor Embedding is a nonlinear technique commonly used to
visualize high-dimensional data in two or three dimensions.
29. What is UMAP?
Uniform Manifold Approximation and Projection is a nonlinear dimensionality reduction
technique used for visualization and other representation-learning tasks.
30. What is NumPy used for?
NumPy supports numerical computation, multidimensional arrays, and mathematical
operations.
31. What is Pandas used for?
Pandas is used to load, clean, manipulate, and analyze tabular data.
32. What is Matplotlib?
It is a Python library for creating graphs and visualizations.
33. What is Seaborn?
It is a statistical visualization library built on Matplotlib.
34. What are TensorFlow and Keras?
TensorFlow is a machine learning framework, and Keras provides a high-level API for
building and training neural networks.
35. What are Kaggle and the UCI Machine Learning Repository?
They are online resources for finding datasets used in ML learning, experimentation, and
research.
36. What is handwritten digit classification?
It is a classification task in which an ML model identifies digits from images, commonly
using the MNIST dataset containing digits 0–9.
37. What is regression in ML?
Regression is a supervised learning technique used to predict continuous numerical values.
38. Give real-world applications of regression.
House price prediction, temperature forecasting, sales prediction, and stock price estimation.
39. Name different types of regression models.
Linear regression, multiple linear regression, polynomial regression, Ridge regression, Lasso
regression, and Elastic Net regression.
40. What is simple linear regression?
It models the relationship between one independent variable and one dependent variable
using a straight line.
41. What is multiple linear regression?
It predicts a numerical output using two or more independent variables.

42. What is the dependent variable?
It is the target variable that the model attempts to predict.
43. What is the independent variable?
It is an input feature used to predict the target variable.
44. What is the purpose of a train-test split?
To train a model on one portion of the data and evaluate generalization on a separate portion.
45. What is MAE?
Mean Absolute Error is the average absolute difference between actual and predicted values.
46. What is MSE?
Mean Squared Error is the average squared difference between actual and predicted values. It
penalizes larger errors more heavily.
47. What is RMSE?
Root Mean Squared Error is the square root of MSE. It expresses prediction error in the same
units as the target variable.
48. What is the R2 score?
R2 measures how well a regression model explains the variation in the target relative to a
baseline that predicts the target mean.
49. Which is better: MAE or MSE?
It depends on the problem. MAE is less sensitive to large errors, while MSE penalizes large
errors more strongly.
50. Can the R2 score be negative?
Yes. A negative R2 means the model performs worse on the evaluated data than predicting the
target mean, using the standard definition.
51. What is regularization?
Regularization adds a penalty to the model's objective to discourage overly complex models
and help reduce overfitting.
52. What is Lasso regression?
Lasso uses L1 regularization and can reduce some feature coefficients to exactly zero.
53. What is Ridge regression?
Ridge uses L2 regularization and shrinks coefficients toward zero without generally making
them exactly zero.
54. What is Elastic Net regression?
Elastic Net combines L1 and L2 regularization.
55. What is the difference between Lasso and Ridge?
Lasso can perform feature selection by setting coefficients to zero; Ridge generally retains all
features but shrinks their coefficients.

56. Why is feature scaling important for regularized regression?
Scaling helps ensure that features are penalized comparably rather than allowing their
measurement units to dominate the penalty.
57. How can regression be used for stock market prediction?
Historical prices and other relevant features can be used to predict a future numerical target,
such as the next day's closing price. Such predictions are uncertain.
58. Why should stock data be split chronologically?
To avoid training on future information and evaluating on past observations, which can cause
data leakage.
59. What is classification?
Classification is a supervised learning task that predicts discrete categories or class labels.
60. What is binary classification?
It predicts one of two classes, such as spam or not spam.
61. What is multiclass classification?
It predicts one class from more than two mutually exclusive classes, such as digits 0–9.
62. What is multilabel classification?
It allows one example to belong to multiple classes simultaneously, such as an image labeled
both “person” and “bicycle.”
63. What is K-Nearest Neighbors (KNN)?
KNN predicts a class using the labels of the nearest training examples.
64. What is the role of K in KNN?
K determines how many neighboring examples are considered when making a prediction.
65. Why is feature scaling important in KNN?
KNN often uses distance measures, so features with larger numerical scales can dominate the
distance if data is not scaled appropriately.
66. What is the Naive Bayes classifier?
It is a probabilistic classifier based on Bayes' theorem that commonly assumes features are
conditionally independent given the class.
67. What is a decision tree?
A decision tree makes predictions using a sequence of feature-based decisions arranged in a
tree structure.
68. What is a leaf node in a decision tree?
It is a terminal node that provides the predicted class or numerical output.
69. What is Support Vector Machine (SVM)?
SVM is an algorithm that finds a decision boundary with a large margin between classes.
Kernel methods can model nonlinear boundaries.

70. What is a kernel in SVM?
A kernel function enables SVM to model relationships corresponding to higher-dimensional
feature spaces without explicitly computing all transformed features.
71. What is a confusion matrix?
It is a table comparing actual and predicted classes, showing correct and incorrect
predictions.
72. What are TP, TN, FP, and FN?
• TP: True Positive
• TN: True Negative
• FP: False Positive
• FN: False Negative
73. What is accuracy?
The proportion of all predictions that are correct.
74. What is precision?
The proportion of predicted positive cases that are actually positive.
75. What is recall?
The proportion of actual positive cases correctly identified by the model.
76. What is the F1-score?
It is the harmonic mean of precision and recall.
77. Why can accuracy be misleading for imbalanced data?
A model may predict the majority class most of the time and still achieve high accuracy while
missing many minority-class cases.
78. Which metric is important in disease prediction?
Recall may be especially important when missing a disease case is costly, but precision,
specificity, and other clinical considerations should also be evaluated.
79. What is a false negative in disease prediction?
It occurs when a person who actually has the disease is predicted as not having it.
80. What is a false positive in disease prediction?
It occurs when a person who does not have the disease is predicted as having it.
81. What is unsupervised learning?
It is a type of ML that identifies patterns or structures in unlabeled data.
82. What is the difference between supervised and unsupervised learning?
Supervised learning uses labeled data, while unsupervised learning works with unlabeled
data.

83. What is clustering?
Clustering groups data points based on their similarity.
84. What is feature selection?
Feature selection identifies relevant input features and removes irrelevant or redundant ones.
85. What is market basket analysis?
It identifies relationships between items frequently purchased together.
86. What is anomaly detection?
It identifies unusual data points that differ significantly from expected patterns.
87. What is document clustering?
It groups similar documents based on their content or feature representations.
88. What is density-based clustering?
It groups closely packed data points into clusters and identifies points in low-density regions
as noise or outliers.
89. What is K-Means clustering?
K-Means partitions data into K clusters by assigning points to the nearest centroid and
updating centroids iteratively.
90. What is a centroid?
It is the mean position of the data points assigned to a cluster in K-Means.
91. How do you select K in K-Means?
Common methods include the Elbow method and Silhouette score, supported by domain
knowledge.
92. What is hierarchical clustering?
It creates a hierarchy of clusters, often represented using a dendrogram.
93. What is a dendrogram?
A tree-like diagram showing how clusters are merged or divided during hierarchical
clustering.
94. What is the Apriori algorithm?
Apriori discovers frequent itemsets and generates association rules from transactional data.
95. What is minimum support?
It is the minimum proportion of transactions that must contain an itemset for it to qualify as
frequent.
96. What is confidence in association rules?
It measures how frequently the consequent appears in transactions that contain the
antecedent.

97. What is lift?
Lift compares the observed co-occurrence of two itemsets with the co-occurrence expected if
they were independent. A lift greater than 1 indicates positive association.
98. What is DBSCAN?
Density-Based Spatial Clustering of Applications with Noise groups dense regions and
identifies low-density points as noise.
99. What are eps and min_samples in DBSCAN?
eps defines the neighborhood radius, while min_samples specifies the minimum number of
points needed in a neighborhood to qualify as a core point.
100. What is the Silhouette score?
It measures how well a point fits within its own cluster compared with other clusters. Values
range from −1 to 1, with higher values generally indicating better-separated clusters.
101. What is the Davies–Bouldin Index?
It measures cluster similarity based on within-cluster dispersion and between-cluster
separation. Lower values are generally better.
102. What is WCSS?
Within-Cluster Sum of Squares measures the sum of squared distances between points and
their assigned cluster centroids.
103. What is the Elbow method?
It plots WCSS against the number of clusters and looks for a point where further reductions
become less substantial.
104. How is ML used for retail customer segmentation?
Clustering algorithms group customers using features such as annual income, spending score,
and purchase behavior.
105. Why is feature scaling important in clustering?
It prevents features with larger numerical ranges from dominating distance-based clustering.
106. What is ensemble learning?
Ensemble learning combines predictions from multiple models to improve predictive
performance or robustness.
107. Why do we use ensemble learning?
It can improve accuracy, reduce variance, and sometimes reduce bias compared with
individual models.
108. What is a homogeneous ensemble?
It combines multiple models of the same algorithm family, such as several decision trees.
109. What is a heterogeneous ensemble?
It combines different types of models, such as decision trees, SVM, and logistic regression.

110. What are the limitations of ensemble learning?
It can increase computational cost, memory requirements, model complexity, and difficulty of
interpretation.
111. What is a voting ensemble?
It combines predictions from multiple models to make a final prediction.
112. What is hard voting?
It selects the class receiving the most votes from the individual classifiers.
113. What is soft voting?
It combines predicted class probabilities and selects the class with the highest average
probability.
114. What is weighted voting?
It assigns different weights to model predictions based on their importance or validated
performance.
115. What is averaging in ensemble learning?
It combines numerical predictions from multiple models, commonly by calculating their
mean.
116. What is bagging?
Bagging trains models on bootstrap samples of the training data and aggregates their
predictions.
117. What is bootstrapping?
It creates new samples by randomly selecting observations from the original dataset with
replacement.
118. What is boosting?
Boosting builds models sequentially, with later models focusing on errors or residual patterns
left by earlier models.
119. What is AdaBoost?
Adaptive Boosting increases the emphasis on previously misclassified training examples so
subsequent learners can focus on them.
120. What is Gradient Boosting?
It builds models sequentially, with each new model fitting the negative gradient of a specified
loss function.
121. What is XGBoost?
Extreme Gradient Boosting is an optimized gradient-boosting implementation that includes
regularization and computational enhancements.
122. What is stacking?

Stacking combines predictions from multiple base models using another model called a meta-
model.

123. What is blending?
Blending trains a meta-model using predictions from base models on a separate holdout
dataset.
124. What is Random Forest?
Random Forest combines many decision trees trained using bootstrap samples and
randomized feature selection.
125. How does Random Forest reduce overfitting?
It averages predictions from diverse trees, which often reduces variance compared with a
single decision tree.
126. What is the difference between bagging and boosting?
Bagging generally trains models independently on bootstrap samples, whereas boosting trains
models sequentially to improve performance on remaining errors.
127. What is the difference between Random Forest and Gradient Boosting?
Random Forest uses a collection of randomized trees, while Gradient Boosting builds trees
sequentially to optimize a loss function.
128. How is ensemble learning used in fraud detection?
Ensemble models can combine multiple patterns and classifiers to identify potentially
fraudulent transactions.
129. What is Reinforcement Learning (RL)?
RL is a learning approach in which an agent learns a policy through interaction with an
environment to maximize cumulative reward.
130. How does RL differ from supervised learning?
Supervised learning uses labeled examples, while RL learns from rewards received through
interaction.
131. What is the multi-arm bandit problem?
It is a decision-making problem in which an agent chooses among actions with uncertain
rewards to maximize cumulative reward.
132. What is exploration?
Exploration means trying actions to gather information about their rewards.
133. What is exploitation?
Exploitation means choosing an action based on current knowledge to obtain a high expected
reward.
134. What is the exploration-exploitation trade-off?
It is the challenge of balancing information gathering with choosing actions believed to yield
higher rewards.
135. What is an agent?
The agent is the learner or decision-maker.

136. What is an environment?
The environment is the system with which the agent interacts.
137. What is a state?
A state represents the relevant situation of the environment at a particular time.
138. What is an action?
An action is a decision taken by the agent in a state.
139. What is a reward?
A reward is a numerical feedback signal indicating the immediate outcome of an action.
140. What is a policy?
A policy specifies how an agent selects actions based on its state.
141. What is return in RL?
Return is the cumulative reward collected over time, often with discounting.
142. What is the discount factor?
The discount factor, \(\gamma\), determines how strongly future rewards contribute to the
return. It typically lies between 0 and 1.
143. What is a value function?
A value function estimates the expected return from a state or state-action pair.
144. What is an MDP?
A Markov Decision Process is a mathematical framework defined by states, actions,
transition probabilities, rewards, and a discount factor.
145. What is the Markov property?
It means the next-state distribution depends on the current state and action, rather than the
complete history, under the chosen state representation.
146. What is the Bellman equation?
It expresses the value of a state or state-action pair in terms of immediate reward and
expected future value.
147. What is model-based learning?
It uses a model of the environment's dynamics and rewards to plan or select actions.
148. What is Temporal Difference (TD) learning?
TD learning updates value estimates using observed rewards and estimates of future values
without waiting for the complete episode to finish.
149. What is generalization in RL?
It is the ability to use learned knowledge to make good decisions in previously unseen states
or situations.

150. What is Q-Learning?
Q-Learning is an off-policy, model-free RL algorithm that learns action values using the best
estimated next-state action value.
