Q1.
Define Artificial Intelligence (2 Marks)
Answer:
Artificial Intelligence (AI) is the branch of computer science that aims to create machines that can
perform tasks that would require intelligence if done by humans.
Important Points:
• It includes learning, reasoning, problem-solving, perception, and language understanding.
• Coined by John McCarthy in 1956.
• AI = Human Intelligence + Machine + Algorithm
Q2. Use of Quantifiers in Predicate Logic (2 Marks)
Answer:
Quantifiers are symbols used in predicate logic to express the quantity of specimens in a domain that
satisfy a formula.
Types:
• Universal Quantifier (∀): Asserts that a predicate holds for all values.
• Existential Quantifier (∃): Asserts that a predicate holds for at least one value.
Example: ∀x (Student(x) → Studies(x))
Q3. Operators in Predicate Logic and Their Precedence (2 Marks)
Operators Used:
• ∧ (AND)
• ∨ (OR)
• → (IMPLIES)
• ↔ (IF AND ONLY IF)
• ¬ (NOT)
Precedence (from highest to lowest):
1. ¬ (Negation)
2. ∧ (Conjunction)
3. ∨ (Disjunction)
4. →, ↔ (Implication and Biconditional)
5. Quantifiers (∀, ∃)
Q4. Unification Algorithm (5 Marks)
Answer:
Unification is a process of making two logical expressions identical by finding a substitution for
variables.
Steps in Unification Algorithm:
1. Compare the expressions.
2. Identify the substitutions to make them identical.
3. Apply substitutions recursively.
4. If no conflict, return substitution; else fail.
Example:
Unify P(x, a) and P(b, y) → Substitution {x/b, y/a}
Q5. Steps of Resolution in Predicate Logic (5 Marks)
Answer:
Resolution is an inference rule used for automated theorem proving.
Steps:
1. Convert all statements to clause form (CNF).
2. Negate the goal and add to the KB.
3. Resolve two clauses to eliminate a literal.
4. Continue till empty clause is derived (i.e., proof complete).
Example:
Given:
• ∀x (¬P(x) ∨ Q(x))
• ∀x (P(a))
Resolve to get Q(a)
Q6. Significance of Predicate Calculus in Knowledge Representation (5 Marks)
Answer:
Predicate Calculus allows expressing facts and rules about the world logically.
Importance:
• Express relationships and quantifiers.
• Enables automated reasoning.
• Provides a foundation for AI applications like expert systems, NLP, etc.
• More expressive than propositional logic.
Q7. Comparison of Knowledge Representation Techniques (5 Marks)
Answer:
Technique Reasoning Type Inference Ability Example Use Case
Semantic Networks Graph-based Hierarchical inference Taxonomies, ontologies
Frames Object-based Slot-based inheritance Expert systems
Predicate Logic Formal logic Deductive reasoning Rule-based systems
Production Rules IF-THEN reasoning Forward/backward chaining Expert systems
Conclusion: Each has strengths. Predicate logic is good for formal inference, while semantic networks
are intuitive for conceptual hierarchy.
Q8. Represent Knowledge in Predicate Logic + Inference (5 Marks)
Statement:
"All students who study AI pass the exam. John is a student who studies AI."
Representation:
1. ∀x (Student(x) ∧ StudiesAI(x) → Passes(x))
2. Student(John) ∧ StudiesAI(John)
Inference:
From (1) and (2) ⇒ Passes(John)
→ Therefore, John will pass the exam
Q10. What is an "ISA" relationship in a semantic network? (2 Marks)
Already a 2-mark question, but here’s a better descriptive version:
Answer:
In a semantic network, the "ISA" relationship is used to represent a hierarchical relationship
between concepts. It means "is a kind of".
• Example: Dog ISA Mammal – means "Dog is a type of Mammal".
• It helps in inheritance: if Mammals have hair, then Dog inherits that property.
• ISA links support structured knowledge and make reasoning easier in AI.
Q11. Define Frames in AI (2 Marks)
Answer:
A Frame is a data structure for dividing knowledge into substructures by representing "stereotyped
situations".
• Each frame consists of slots (attributes) and slot values.
• It’s used in knowledge representation.
• Example: A "Car" frame may have slots like make, model, engine, color.
Q12. Construct a Semantic Network for: "Dog", "Mammal", "Animal", "Has Fur", "Barks" (5
Marks)
Answer:
A semantic network is a graphical representation of knowledge using nodes (concepts) and links
(relationships).
Nodes:
• Dog
• Mammal
• Animal
• Has Fur
• Barks
Relationships:
• Dog ISA Mammal
• Mammal ISA Animal
• Dog → Has Fur
• Dog → Barks
Explanation:
• Dog is a subclass of Mammal.
• Mammal is a subclass of Animal.
• Dog possesses the properties: Has Fur and Barks.
Diagram (for exam):
Draw a semantic network like this:
less
Copy code
[Dog] ---ISA---> [Mammal] ---ISA---> [Animal]
| |
| V
+---> [Has Fur] [Barks]
Q14. Pipeline of Natural Language Processing (5 Marks)
Answer:
The NLP pipeline refers to the sequence of steps through which natural language data is processed in
an AI system.
1. Text Preprocessing
• Tokenization: Splitting sentences into words.
• Stop-word removal: Removing common words like "is", "the".
• Stemming/Lemmatization: Reducing words to base form (e.g., "running" → "run").
2. Part-of-Speech Tagging
• Assigning grammatical tags like noun, verb, adjective to words.
3. Named Entity Recognition (NER)
• Identifying entities such as names, dates, places in text.
4. Syntax Analysis (Parsing)
• Analyzing grammar structure of sentences (parse trees, dependency trees).
5. Semantic Analysis
• Understanding the meaning and context of the text.
6. Sentiment Analysis / Pragmatic Analysis
• Deriving subjective information like emotions or intentions.
Example Flow:
Input: "John bought a red car."
→ Tokenization → POS Tagging → Parsing → NER → Semantic interpretation.
Q15. Construct a Parse Tree for: "Every Dog has bitten the Postmen" (5 Marks)
Answer:
A parse tree is a tree that represents the syntactic structure of a sentence according to a grammar.
Sentence:
"Every Dog has bitten the Postmen"
Steps:
• Every Dog → NP (Noun Phrase)
• has bitten → VP (Verb Phrase)
• the Postmen → NP
Parse Tree Structure:
mathematica
Copy code
S
/ \
NP VP
/\ /\
Det N Aux VP
(Every)(Dog)(has)(V NP)
/ \
(bitten)(the Postmen)
Explanation:
• Sentence (S) = NP + VP
• NP = Every + Dog
• VP = has + bitten + NP
• NP (object) = the + Postmen
17) Explain in Detail about Architecture of Expert System Shell with Block Diagram (5 Marks)
Expert System Shell Architecture:
An Expert System Shell is a software development environment designed to build expert systems. It
provides the core components (inference engine, knowledge base, user interface) without containing
domain-specific knowledge.
Block Diagram:
sql
Copy code
+-------------------+
| User Interface | <-- User queries and inputs
+-------------------+
+-------------------+ +--------------------+
| Inference Engine |<--->| Knowledge Base |
+-------------------+ +--------------------+
| ^
v |
+-------------------+ +--------------------+
| Explanation Subsy.| | Knowledge Acquisition |
+-------------------+ +--------------------+
+-------------------+
| Working Memory |
+-------------------+
Components:
• User Interface: Allows users to interact with the system.
• Inference Engine: Core component that applies logical rules to the knowledge base to
deduce new information.
• Knowledge Base: Stores domain-specific facts and rules.
• Explanation Subsystem: Justifies the reasoning and conclusions.
• Knowledge Acquisition Module: Facilitates entering new rules and knowledge.
• Working Memory: Temporary storage for facts and intermediary resu
18) Knowledge Engineering is the process of creating systems that simulate the decision-making
ability of a human expert. It involves acquiring expert knowledge, organizing it, encoding it into a
usable form, and validating it within an expert system.
Key Steps:
1. Knowledge Acquisition – Gathering knowledge from experts or data.
2. Knowledge Representation – Structuring knowledge using rules, frames, semantic networks,
etc.
3. Knowledge Validation – Ensuring accuracy and consistency.
4. Knowledge Refinement – Improving the knowledge base over time.
19 ) What is the Certainty Factor and how is it used in Expert Systems? (2 Marks)
Certainty Factor (CF) represents the degree of belief or disbelief in a particular piece of information,
ranging from -1.0 (completely false) to +1.0 (completely true).
Usage:
• Helps expert systems handle uncertainty.
• Each rule or fact is assigned a CF value.
• During reasoning, CFs are combined (using mathematical formulas) to infer conclusions with
varying degrees of confidence.
20) Explain how an Augmented Transition Network (ATN) can be used to parse the sentence: "The
cat sat on the mat." (5 Marks)
ATN (Augmented Transition Network):
ATN is a graph-based formalism used for natural language parsing. It represents grammatical rules as
finite state machines.
Parsing Steps for the sentence:
Sentence: "The cat sat on the mat."
• S → NP VP
• NP → Det + Noun → The + cat
• VP → Verb + PP → sat + (on the mat)
ATN States:
1. Start at Sentence Node.
2. Move to NP Node, parse "The cat".
3. Transition to VP Node, parse "sat".
4. Recognize prepositional phrase "on the mat".
5. Parse complete.
Use in NLP:
• Validates grammar structure.
• Converts natural language into parse trees or meaning representations.
Q21. A hospital is designing an expert system for drug prescription. Apply the architecture of
expert systems to define the main components and their roles. (5 Marks)
Designing an expert system for drug prescription involves integrating medical expertise with
decision-making capabilities to assist doctors in recommending safe and effective medications.
Application of Expert System Architecture:
1. User Interface:
o Doctors or medical staff interact with the system through a simple UI.
o Inputs include patient symptoms, age, weight, allergies, medical history, and lab
results.
2. Knowledge Base:
o Contains medical rules, drug information, contraindications, side effects, dosage
guidelines, and treatment protocols.
o Example rule:
If the patient has high blood pressure and is over 65, avoid NSAIDs.
3. Inference Engine:
o Analyzes user input (e.g., patient data) against rules in the knowledge base.
o Deduces appropriate prescriptions or raises warnings for unsafe drug combinations.
o Supports decision-making through forward chaining.
4. Working Memory:
o Temporarily holds patient-specific data for each session.
o This data is used by the inference engine during rule evaluation.
5. Explanation Subsystem:
o Justifies the system’s recommendations.
o Example: "Paracetamol was recommended instead of ibuprofen due to patient's
renal history."
6. Knowledge Acquisition Module:
o Allows pharmacists or medical experts to update drug data or rules.
o Facilitates continuous improvement based on medical advancements.
Conclusion:
By adopting the expert system architecture, hospitals can enhance drug prescription safety, reduce
human errors, and personalize treatment, ultimately improving patient care.
Q22. Analyze the key phases in the development cycle of an expert system and explain how
knowledge acquisition plays a critical role in its success. (5 Marks)
Developing a robust expert system requires a structured approach to ensure it mimics expert-level
decision-making. The development cycle consists of several phases, with knowledge acquisition
being the most critical.
Key Phases in the Expert System Development Cycle:
1. Problem Identification:
o Define the problem domain and scope.
o Understand user needs and specify goals of the expert system.
2. Knowledge Acquisition:
o This is the process of gathering information from domain experts, documents,
databases, or sensors.
o Methods include interviews, questionnaires, case analysis, and observation.
o Tools like knowledge editors and AI models assist in this process.
o Importance: The quality of the expert system directly depends on the accuracy,
completeness, and consistency of the acquired knowledge.
3. Knowledge Representation:
o Organize acquired knowledge using rules (IF-THEN), semantic networks, frames, or
ontologies.
o Ensure knowledge is machine-readable and logically structured.
4. System Design and Implementation:
o Choose an appropriate expert system shell or develop a custom engine.
o Integrate knowledge base and inference mechanisms.
5. Testing and Validation:
o Verify the system's accuracy by comparing its decisions to those of human experts.
o Identify gaps or inconsistencies.
6. Deployment and Maintenance:
o Deploy the system for real-world use.
o Continuously update the system as new knowledge emerges or domain
requirements change.
Conclusion:
Among all phases, knowledge acquisition is the foundation of success. A poor acquisition process
leads to weak reasoning, flawed decisions, and reduced reliability of the expert system. Thus,
capturing expert knowledge accurately and comprehensively is essential.
Q23. Compare and contrast expert system shells with traditional expert system development.
Why are expert system shells preferred for rapid prototyping? (5 Marks)
Expert systems can be developed using two main approaches: traditional development (from
scratch) and expert system shells (pre-built frameworks). Each has its own pros and cons, but shells
are often favored for rapid and efficient development.
Comparison Table:
Feature Traditional Expert System Expert System Shell
Development Time Long (from scratch) Short (prebuilt components)
Programming Skill High Moderate or low
Reusability Low High (modular components)
Flexibility Very flexible, but complex Flexible within framework limits
Feature Traditional Expert System Expert System Shell
Cost and Maintenance Expensive and complex Cost-effective and manageable
Knowledge Separation Often mixed with logic Clearly separated from logic
Why Expert System Shells Are Preferred for Rapid Prototyping:
1. Prebuilt Inference Engines:
o Saves time and effort; only domain-specific knowledge needs to be added.
2. User-Friendly Interfaces:
o Non-programmers (like domain experts) can input knowledge using GUIs.
3. Faster Testing and Iteration:
o Easy to test, modify, and update rules or logic during development.
4. Support for Standard Features:
o Shells come with built-in support for uncertainty management (e.g., certainty
factors), explanation, and rule chaining.
5. Cost-effective:
o Reduces development costs by eliminating redundant work.
Conclusion:
Expert system shells accelerate development, support rapid prototyping, and are ideal for iterative,
evolving projects. They are especially beneficial in real-world applications where expert knowledge
frequently changes.
Q24. Explain the Importance of Expert System Shell in Detail (5 Marks)
An Expert System Shell plays a pivotal role in simplifying the development of expert systems by
providing a generic structure for reasoning and decision-making, without tying it to a specific
domain.
Importance of Expert System Shells:
1. Accelerates Development:
o Developers only need to focus on the domain knowledge (rules, facts).
o No need to build inference mechanisms from scratch.
2. Standardization:
o Offers a consistent structure and methodology for expert system design.
o Reduces errors and increases system reliability.
3. Supports Complex Reasoning:
o Built-in inference engines support both forward and backward chaining.
o Handles uncertainties using certainty factors, fuzzy logic, etc.
4. Ease of Maintenance and Updates:
o Rules and knowledge can be updated without altering the core engine.
o Encourages modularity and separation of concerns.
5. Non-programmer Friendly:
o Knowledge engineers or domain experts with minimal programming skills can input
knowledge directly through user-friendly interfaces.
6. Rapid Prototyping and Experimentation:
o Encourages trial-and-error development.
o Especially useful in domains like healthcare, finance, and diagnostics where frequent
updates are required.
Q25. Write a Note on Dempster-Shafer Theory (5 Marks)
The Dempster-Shafer Theory (DST), also known as the Theory of Belief Functions, is a mathematical
theory of evidence used to model epistemic uncertainty. It generalizes Bayesian probability and
allows reasoning with partial and incomplete information.
Key Concepts:
1. Frame of Discernment (Θ):
o The set of all possible hypotheses. For example, Θ = {Disease A, Disease B}.
2. Basic Probability Assignment (BPA):
o Assigns belief mass (between 0 and 1) to subsets of Θ.
o Total mass must sum to 1.
o For example, m({A}) = 0.6, m({B}) = 0.3, m({A, B}) = 0.1
3. Belief Function (Bel):
o Represents the degree of certainty or evidence in favor of a proposition.
o It is the sum of all masses supporting a subset.
4. Plausibility Function (Pl):
o Measures how much a proposition is not disbelieved.
o Pl(A) = 1 – Bel(not A)
5. Dempster’s Rule of Combination:
o Combines evidence from different sources.
o Helps update belief when new data is available.
Importance:
• Suitable when information is incomplete or uncertain.
• Widely used in sensor fusion, diagnostics, decision support systems, and expert systems.
Conclusion:
Dempster-Shafer Theory provides a flexible and powerful framework for handling uncertain
knowledge, especially when probabilities are hard to define precisely.
Q27. Infer Fuzzy Logic with its Advantages and Disadvantages (5 Marks)
Fuzzy Logic is a form of many-valued logic that deals with reasoning that is approximate rather than
fixed and exact. Unlike classical logic (which uses binary true/false values), fuzzy logic allows values
between 0 and 1.
Key Concepts:
• Fuzzy logic is based on fuzzy sets, where an element’s membership is expressed in degrees
(e.g., 0.6, 0.8).
• A fuzzy system contains: fuzzifier, rule base, inference engine, and defuzzifier.
Advantages:
1. Deals with Uncertainty:
o Ideal for real-world problems where exact data isn’t available.
2. Human-like Reasoning:
o Mimics decision-making like humans using linguistic variables (e.g., "hot", "cold").
3. Robust to Noise:
o Performs well even with noisy or imprecise inputs.
4. Simplicity:
o Easier to implement for control systems (e.g., washing machines, air conditioners).
Disadvantages:
1. Rule Explosion:
o Complex systems require many rules, which become hard to manage.
2. Lack of Learning:
o Pure fuzzy systems don’t learn; they must be manually tuned unless combined with
AI.
3. Interpretability:
o The rules and outputs can sometimes be hard to interpret.
Conclusion:
Fuzzy logic is a practical tool in systems that require approximate reasoning, especially where binary
logic fails, though it must be carefully designed to avoid complexity.
Q28. Compare Simple and Conditional Probability (5 Marks)
Feature Simple Probability Conditional Probability
Probability of a single event Probability of an event given that another event
Definition
happening has occurred
Notation P(A) P(A
P(A) = (favorable outcomes) / (total
Formula P(A
outcomes)
Dependency Independent (standalone) event Depends on occurrence of another event
Example P(rolling a 4) on a die = 1/6 P(rain
Conclusion:
Simple probability is used for standalone events, while conditional probability is critical when events
are interrelated, as it considers additional known information.
Q30. Explain Bayes’ Theorem with Example (5 Marks)
Bayes’ Theorem provides a way to update prior beliefs based on new evidence. It relates conditional
and marginal probabilities of events.
Formula:
P(A∣B)=P(B∣A)⋅P(A)P(B)P(A|B) = \frac{P(B|A) \cdot P(A)}{P(B)}P(A∣B)=P(B)P(B∣A)⋅P(A)
Where:
• P(A∣B)P(A|B)P(A∣B) = Posterior probability (A given B)
• P(B∣A)P(B|A)P(B∣A) = Likelihood (B given A)
• P(A)P(A)P(A) = Prior probability of A
• P(B)P(B)P(B) = Marginal probability of B
Example:
A medical test for a disease is 98% accurate. The disease occurs in 1 out of 1000 people.
Let:
• A = person has disease → P(A) = 0.001
• B = test is positive
Given:
• P(B|A) = 0.98 (true positive rate)
• P(B|¬A) = 0.02 (false positive rate)
Now,
P(B)=P(B∣A)P(A)+P(B∣¬A)P(¬A)=(0.98)(0.001)+(0.02)(0.999)=0.00098+0.01998=0.02096P(B) =
P(B|A)P(A) + P(B|¬A)P(¬A) = (0.98)(0.001) + (0.02)(0.999) = 0.00098 + 0.01998 =
0.02096P(B)=P(B∣A)P(A)+P(B∣¬A)P(¬A)=(0.98)(0.001)+(0.02)(0.999)=0.00098+0.01998=0.02096
P(A∣B)=0.000980.02096≈0.0467=4.67%P(A|B) = \frac{0.00098}{0.02096} ≈ 0.0467 =
4.67\%P(A∣B)=0.020960.00098≈0.0467=4.67%
Interpretation:
Even if the test is positive, there's only a 4.67% chance the person actually has the disease due to low
base rate.
Q31: Bayes’ Theorem Application – Probability Defective Bulb is from Machine A
Given:
• Machine A produces 60% of bulbs → P(A) = 0.6
• Machine B produces 40% of bulbs → P(B) = 0.4
• Defective rate for A → P(D|A) = 0.05
• Defective rate for B → P(D|B) = 0.10
We are asked to find:
P(A|D) – the probability that a defective bulb is from Machine A.
Bayes’ Theorem Formula:
P(A∣D)=P(D∣A)⋅P(A)P(D)P(A|D) = \frac{P(D|A) \cdot P(A)}{P(D)}P(A∣D)=P(D)P(D∣A)⋅P(A)
Step 1: Find Total Probability of Defective Bulb (P(D))
P(D)=P(D∣A)⋅P(A)+P(D∣B)⋅P(B)=(0.05×0.6)+(0.10×0.4)=0.03+0.04=0.07P(D) = P(D|A) \cdot P(A) +
P(D|B) \cdot P(B) \\ = (0.05 \times 0.6) + (0.10 \times 0.4) = 0.03 + 0.04 =
0.07P(D)=P(D∣A)⋅P(A)+P(D∣B)⋅P(B)=(0.05×0.6)+(0.10×0.4)=0.03+0.04=0.07
Step 2: Apply Bayes’ Theorem
P(A∣D)=0.05×0.60.07=0.030.07≈0.4286P(A|D) = \frac{0.05 \times 0.6}{0.07} = \frac{0.03}{0.07} ≈
0.4286P(A∣D)=0.070.05×0.6=0.070.03≈0.4286
Final Answer:
The probability that a randomly selected defective bulb was produced by Machine A is 42.86%.
Q32: Fuzzy Logic – Washing Time (Defuzzification Using Weighted Average)
Given:
• Low dirt → 15 min
• Medium dirt → 30 min, with 70% membership → μ(Medium) = 0.7
• High dirt → 45 min, with 30% membership → μ(High) = 0.3
We use the Weighted Average (Centroid) Method:
Washing Time=∑(μi⋅ti)∑μi\text{Washing Time} = \frac{\sum (\mu_i \cdot t_i)}{\sum
\mu_i}Washing Time=∑μi∑(μi⋅ti) =(0.7⋅30)+(0.3⋅45)0.7+0.3=21+13.51.0=34.5= \frac{(0.7 \cdot 30) +
(0.3 \cdot 45)}{0.7 + 0.3} = \frac{21 + 13.5}{1.0} = 34.5=0.7+0.3(0.7⋅30)+(0.3⋅45)=1.021+13.5=34.5
Final Answer:
The defuzzified washing time is 34.5 minutes.
Q33: α-Cut (0.5-Cut) of Fuzzy Set
Fuzzy Set A:
A = {(10, 1.0), (15, 0.8), (20, 0.5), (25, 0.2), (30, 0.0)}
Definition:
An α-cut includes all elements of a fuzzy set whose membership degree is greater than or equal to
α.
For α = 0.5, we take all elements where μ(x) ≥ 0.5.
From the set:
• (10, 1.0)
• (15, 0.8)
• (20, 0.5)
• (25, 0.2)
• (30, 0.0)
Final Answer:
0.5-cut of A = {10, 15, 20}
Q34: Explain the Perspective of Machine Learning
Machine Learning (ML) is the scientific study of algorithms that enable computers to learn from data
and make predictions or decisions without being explicitly programmed.
Key Perspectives:
1. Learning from Data:
ML uses data to learn patterns and relationships, allowing the system to predict outcomes.
2. Improvement Over Time:
ML systems improve their performance with more data and experiences (feedback).
3. Model-Driven Intelligence:
ML constructs models to generalize insights from past data and apply it to new cases.
4. Applications:
Used in image recognition, language translation, spam detection, medical diagnosis, and
many other domains.
Final Answer:
ML is a core field of AI that emphasizes data-driven learning, pattern discovery, and self-
improvement through experience, making it essential for intelligent systems.
Q35: Factors to Consider When Designing an ML System
1. Data Availability & Quality:
Good quality, relevant, and sufficient data is essential for accurate model training.
2. Problem Definition:
Clearly define whether the problem is classification, regression, clustering, etc.
3. Model Selection:
Choose appropriate ML algorithms (e.g., SVM, Decision Trees, Neural Networks) suited for
the task.
4. Feature Engineering:
Transform raw data into meaningful features to enhance model performance.
5. Evaluation Metrics:
Use precision, recall, accuracy, F1-score depending on the use case.
6. Scalability & Deployment:
Ensure the model works in real-world scenarios and can be maintained over time.
Final Answer:
Designing a successful ML system involves careful data handling, model choice, evaluation, and
deployment readiness to ensure effective performance.
Q36: Diabetes Prediction – Type of ML Problem
Given Problem:
Predicting whether a patient has diabetes based on medical history.
ML Perspective:
• Type of Learning: Supervised Learning
• Task: Binary Classification (Diabetic or Not)
• Input: Medical features (glucose level, age, BMI, etc.)
• Output: Diabetes diagnosis (Yes/No)
Justification:
• Historical data is labeled with diabetes outcomes.
• Model learns patterns from labeled training data.
• Classification algorithms like Logistic Regression, Random Forest, or SVM can be used.
Final Answer:
This is a supervised classification problem where the model predicts diabetes using labeled
historical data.
Q37: Issues in Machine Learning
1. Overfitting:
Model learns noise in training data and performs poorly on new data.
2. Underfitting:
Model is too simple and fails to learn patterns.
3. Data Bias & Quality:
Biased or poor-quality data leads to unfair or inaccurate outcomes.
4. Interpretability:
Complex models like deep neural networks are hard to explain.
5. Computational Cost:
Training large models needs high computational power and time.
6. Ethical Concerns:
ML decisions (e.g., in hiring or loan approval) can reinforce societal biases.
Final Answer:
ML faces challenges like overfitting, biased data, interpretability, and computational cost, which
affect its performance and trustworthiness.
Q38: Real-World Scenario for Reinforcement Learning
Scenario: Autonomous Drone Navigation
Why RL is Suitable:
• The drone learns by interacting with its environment.
• It gets rewards for reaching the destination, avoiding obstacles.
• It receives penalties for crashing or going off-route.
• The drone learns policies over time through trial and error.
Comparison:
• Supervised Learning requires labeled data (hard to label every drone action).
• Unsupervised Learning doesn't provide feedback.
• RL is ideal for environments with sequential decision-making.
Final Answer:
Reinforcement Learning is best for dynamic tasks like drone navigation, where the agent learns
optimal actions via trial and reward feedback.
Q39: End-to-End Machine Learning Pipeline Framework
1. Problem Understanding:
Define whether the problem is classification, regression, or clustering.
2. Data Collection:
Gather raw data from sensors, databases, or APIs.
3. Data Preprocessing:
Handle missing values, normalize data, and remove outliers.
4. Feature Engineering:
Select and construct features that improve model performance.
5. Model Training:
Use algorithms like SVM, Decision Trees, or Neural Networks.
6. Model Evaluation:
Use metrics like accuracy, precision, recall, F1-score.
7. Model Tuning:
Optimize hyperparameters using Grid Search or Cross-validation.
8. Deployment:
Serve the model through APIs or integrated applications.
9. Monitoring & Maintenance:
Continuously track performance and retrain with new data.
Final Answer:
The end-to-end ML pipeline involves data handling, model building, deployment, and real-time
maintenance to create robust and scalable AI systems.
Q40: Ill-Posed vs Well-Posed Learning Problems
Criteria Well-Posed Problem Ill-Posed Problem
Definition Has a unique and stable solution No unique or stable solution exists
Satisfies Hadamard's criteria (existence,
Mathematics Violates at least one Hadamard criterion
uniqueness, stability)
Reconstructing blurred image from
Examples Predicting exam marks using study hours
noisy input
Needs additional constraints or
Solvability Easily solvable with ML models
regularization
Small changes in data → small changes in Small changes in data → large
Stability
output unpredictable changes
Final Answer:
Well-posed problems have clearly defined and stable solutions, while ill-posed problems are
uncertain and require mathematical techniques like regularization to solve effectively.
