Human Computer Interaction

Assignment Questions
[Write the questions and answers using Pen and Paper. Submit it in a channel file on or before
13/02/2026]

1. Critically evaluate the limitation of comparing human learning with machine learning.
Using a real-world Human–Computer Interaction (HCI) example, explain why a
system that learns from data may still fail to understand user intent or context in the
same way a human does.
2. Machines are often described as being “unable to take decisions on their own.”
Critically examine the extent to which supervised learning challenges this claim. In
your answer, distinguish between decision-making and autonomy, and support your
reasoning with an appropriate example.
3. In machine learning, “experience” is often represented in the form of data.
Explain why this abstraction is necessary, and critically discuss how poor or
incomplete data representation can distort the system’s learned experience and lead to
incorrect or biased outcomes.
4. Can a feature that is uninformative or misleading in one problem become critical in
another? Justify your answer using a relevant Human–Computer Interaction (HCI)
example, and explain how task context influences feature usefulness.
5. The Vector Space Model represents user interactions using numerical features.
Critically evaluate how this assumption influences the modeling of subjective HCI
attributes such as comfort, trust, or satisfaction, and discuss the limitations this creates
for intelligent interactive systems.
6. If two features individually provide weak classification performance, under what
conditions can their combination lead to a significant improvement in classification
accuracy? Explain your reasoning with appropriate examples or arguments.
7. Many classifiers assign a label by selecting the class with the highest conditional
probability. Critically discuss why this decision rule may be unreliable or unsafe in
safety-critical Human–Computer Interaction systems, and explain the potential
consequences of such failures.
8. In some classification problems, estimated probabilities take extreme values such as
0% or 100%. Critically analyze what such extreme probabilities indicate about the
size and diversity of the training dataset, and discuss their implications for a model’s
ability to generalize to unseen cases.
9. Class imbalance is a common issue in real-world datasets. Explain how class
imbalance affects probability estimation in Naive Bayes classifiers, and relate this
problem to challenges encountered in collecting and labeling Human–Computer
Interaction data.
 10. Many digital systems provide all required functions, yet are still perceived as poorly
designed. Critically analyze why functional availability alone is insufficient to ensure
good design from a Human–Computer Interaction perspective.
11. Indicators such as confusing navigation, poor feedback, and high error rates are often
treated as surface-level usability problems. Identify the deeper design failures that
give rise to these symptoms, and explain how such failures typically emerge during
the design and development process.
12. Usability is commonly evaluated using dimensions such as efficiency, effectiveness,
learnability, memorability, and satisfaction. Identify a real-world digital product that
performs well in terms of efficiency but poorly in terms of user satisfaction, and
critically analyze the reasons for this mismatch.
13. In minimalist interface design, affordance and visibility can come into conflict.
Discuss this conflict using a relevant example, and argue how a designer should
balance simplicity with usability when resolving it.
14. A system may perform well during controlled usability testing yet fail when deployed
in real-world settings. Identify an evaluation approach that could help prevent such
failures and explain why this approach is effective in capturing real-world user
behavior.
15. The Table 1 represents interaction-level features collected from users performing a
task in a Human–Computer Interaction (HCI) system. Each sample is classified into
either Low cognitive load (C1) or High cognitive load (C2).

(a) Using this dataset, explain how a Naive Bayes classifier would model the
relationship between interaction features (Mouse Speed, Error Rate, Task
Time, Keystroke Pause, and Navigation) and the class labels.
(b) Identify two features from the table that are likely to violate the conditional
independence assumption of Naive Bayes, and justify your choice using
interaction behavior reasoning.
(c) One sample in Class C₂ shows “Efficient” navigation despite high error rate
and long task time. Critically analyze what this inconsistency reveals about
the limitations of feature-based classification in HCI.
(d) If this classifier were used to drive real-time interface adaptation (e.g.,
simplifying the UI for Class C₂ users), discuss one potential risk of
misclassification and its impact on user experience.

Table 1: Training Data

M K
E (Error T (Task N Cognitive
Sample (Mouse (Keystroke
Rate) Time) (Navigation) Load
Speed) Pause)
1 Fast Low Short Short Efficient Low
2 Slow High Short Long Inefficient High
3 Fast High Long Long Inefficient High
4 Fast Low Long Short Efficient Low
5 Slow Low Short Short Efficient Low
6 Fast High Long Long Inefficient High
7 Slow High Long Long Efficient High
8 Slow Low Long Long Inefficient High
9 Slow High Long Long Inefficient High
10 Slow High Long Long Inefficient High
11 Slow High Long Short Inefficient High
12 Slow High Short Long Inefficient High
13 Fast Low Short Short Efficient Low
14 Fast Low Long Short Efficient Low
15 Slow Low Short Long Efficient Low
16 Fast High Long Short Inefficient High
17 Fast Low Short Short Efficient Low
18 Fast Low Short Long Efficient Low
19 Slow High Long Long Inefficient High
20 Fast Low Long Long Inefficient High
Excellent. From **Q4 to Q15**, I’ll now write **10-mark, NIT-style answers**.

Format followed (as expected in NIT exams):

* **Introduction**
* **Conceptual Explanation**
* **Critical Analysis**
* **Relevant HCI Example**
* **Implications**
* **Conclusion (Strong closing line)**

Each answer ≈ 250–300 words (ideal for 10 marks).

---

# **Q4. Feature Usefulness and Task Context**

### Introduction

In machine learning and HCI, feature usefulness is not absolute but context-dependent. A
feature that appears irrelevant in one task may become highly significant in another.

### Conceptual Explanation

A feature is informative if it reduces uncertainty about the target variable. Its importance
depends on:

* Nature of the task

* User goal
* Interaction environment
* Outcome being predicted

### Critical Analysis

Feature relevance is influenced by:

1. **Task objective** – Predicting cognitive load vs predicting typing accuracy.

2. **User behavior variation** – Same feature may represent different meanings.
3. **Environmental factors** – Context alters interpretation of features.

### HCI Example

Mouse speed:

* In typing skill assessment → may be irrelevant.

* In gaming or cognitive load detection → critical indicator.
Similarly, eye-tracking fixation duration:
* Irrelevant in simple data entry.
* Critical in usability testing for confusion detection.

### Implications

Designers must perform task analysis before feature selection. Misidentifying features can
reduce model accuracy and misinterpret user states.

### Conclusion

Feature importance is fundamentally contextual. Without understanding task goals, feature

selection becomes unreliable and misleading.
---

# **Q5. Vector Space Model and Subjective HCI Attributes**

### Introduction

The Vector Space Model (VSM) represents user interactions as numerical feature vectors.
While effective for measurable variables, it faces limitations in modeling subjective
experiences.

### Conceptual Explanation

In VSM:
User interaction → numerical features → vector representation → classification or clustering.

### Critical Analysis

Subjective attributes such as:

* Comfort
* Trust
* Satisfaction
are multidimensional and emotional. Reducing them to numerical values creates problems:

1. **Loss of nuance** – Emotional states cannot be fully quantified.

2. **Ambiguity** – Same numeric pattern may represent different feelings.
3. **Measurement bias** – Self-reported satisfaction scores are inconsistent.

### HCI Example

Fast clicking:
* May indicate efficiency.
* May indicate anxiety.
Numeric features alone cannot differentiate emotional intent.

### Limitations for Intelligent Systems

* Poor emotional inference

* Oversimplification of human behavior
* Reduced personalization accuracy

### Conclusion

While VSM enables computational modeling, it inadequately captures subjective human

experience, limiting intelligent interactive systems.

---

# **Q6. Combining Weak Features for Strong Classification**

### Introduction

Individual features may have weak predictive power, yet their combination can significantly
improve classification accuracy.

### Conceptual Explanation

This occurs when:

* Features are conditionally independent.

* They capture complementary information.
* Their joint probability reduces uncertainty effectively.

### Theoretical Basis

Using Naive Bayes or logistic regression:

Even weak likelihood ratios multiply to produce strong posterior probabilities.

### HCI Example

Predicting high cognitive load:

* Mouse speed alone → weak predictor.

* Error rate alone → weak predictor.
* Together (slow + high error) → strong signal of overload.

### Conditions Required

1. Complementarity
2. Non-redundancy
3. Different behavioral dimensions captured

### Critical Insight

If two weak features are highly correlated, combining them does not improve performance.

### Conclusion

Weak individual predictors can produce strong classification when they provide independent
and complementary evidence.

---
# **Q7. Highest Conditional Probability in Safety-Critical Systems**

### Introduction

Many classifiers use the Maximum A Posteriori (MAP) rule—selecting the class with the
highest probability. However, this approach may be unsafe in critical HCI applications.

### Critical Analysis

Problems with highest probability rule:

1. **Ignores uncertainty margin**

51% vs 49% treated same as 90% vs 10%.
2. **Ignores cost of error**
False negative may be catastrophic.
3. **Overconfidence in imperfect data**
Small datasets may misrepresent reality.

### HCI Example

Medical decision support interface:

If cancer probability = 55%, system labels as “cancer”.
But uncertainty is high. Wrong decision may cause panic or delay treatment.

In aviation cockpit interfaces:

Wrong classification of pilot stress can risk lives.

### Consequences

* Safety hazards
* Ethical concerns
* Loss of trust
* Legal liability

### Conclusion

In safety-critical HCI systems, probability maximization alone is insufficient. Risk-aware and

cost-sensitive decision rules are necessary.

---

# **Q8. Extreme Probability Estimates**

### Introduction

Probabilities of 0% or 100% indicate absolute certainty, which is rarely realistic in real-world

data.

### What They Indicate

1. Small training dataset

2. Lack of feature diversity
3. Overfitting
4. Data sparsity problem

### Critical Analysis

Extreme values suggest that:

* Model has never observed contradictory evidence.

* The training data does not represent full variability.
### HCI Example

If all “High cognitive load” samples show slow mouse speed, the model assigns 100%
probability to slow speed under high load.
In reality, expert users may move fast even under stress.

### Implications

* Poor generalization
* High prediction errors on unseen cases
* Overconfidence in decisions

### Conclusion

Extreme probabilities often signal limited or biased training data, reducing reliability in real-
world deployment.

---

# **Q9. Class Imbalance and Naive Bayes**

### Introduction

Class imbalance occurs when one class significantly outnumbers others. This affects
probability estimation in Naive Bayes classifiers.

### Effect on Naive Bayes

Posterior probability:
[
P(C|X) ∝ P(C) × P(X|C)
]

If P(C) is large (majority class), it dominates the decision even if feature evidence is weak.

### HCI Context

In cognitive load detection:

* Most interactions may be low load.

* High load events are rare.

Thus, classifier tends to predict low load.

### Challenges in HCI Data Collection

* Rare events difficult to capture

* Manual labeling expensive
* Ethical constraints

### Consequences

* Reduced sensitivity to minority class

* Biased interface adaptation
* Poor personalization

### Conclusion

Class imbalance distorts probability estimation and must be addressed using balancing
techniques or cost-sensitive learning.
---

# **Q10. Functional Availability vs Good Design**

### Introduction

Functional completeness does not guarantee usability or good design in HCI.

### Critical Analysis

Good design requires:

* Usability
* Learnability
* Feedback
* Cognitive simplicity

A system may provide all features but:

* Be difficult to navigate
* Overload users cognitively
* Lack proper feedback

### HCI Example

Many government service portals provide complete functionality but are confusing due to
poor layout and unclear instructions.

### Core Insight

Design quality depends on user experience, not just technical capability.

### Conclusion

From an HCI perspective, usability and interaction quality determine success, not mere
functional availability.

---

(Questions 11–15 will continue in same detailed NIT 10-mark format. Reply “Continue” and
I’ll complete the remaining 5 in equal depth.)

Perfect. Continuing in the same **NIT 10-mark structured format** (Introduction →

Conceptual Explanation → Critical Analysis → HCI Example → Implications →
Conclusion).

---

# **Q11. Deeper Design Failures Behind Usability Problems**

### Introduction

Indicators such as confusing navigation, poor feedback, and high error rates are often treated
as surface usability issues. However, these symptoms usually arise from deeper systemic
design failures.

### Conceptual Explanation

Usability problems emerge when there is a mismatch between:

* System model (how designers think it works)
* User mental model (how users expect it to work)

### Critical Analysis

Deeper causes include:

1. **Lack of User-Centered Design (UCD)** – No early user research.

2. **Poor Requirement Analysis** – Functional focus without behavioral understanding.
3. **Technology-Driven Development** – Engineers prioritizing system capability over
usability.
4. **Absence of Iterative Testing** – No prototyping or usability validation.
5. **Organizational Pressure** – Deadlines overriding user research.

### HCI Example

In many e-governance portals, navigation is layered deeply because backend structure

dictates UI design rather than user workflow. This leads to confusion and high task
abandonment.

### Implications

Surface-level fixes (changing button color or text) do not solve fundamental structural issues.

### Conclusion

Usability problems are rarely accidental; they are the result of early-stage design failures and
lack of human-centered thinking.

---

# **Q12. Efficient but Unsatisfying Product**

### Introduction

A product may perform efficiently yet fail to satisfy users emotionally. Efficiency measures
speed and accuracy, while satisfaction measures perceived comfort and trust.

### Conceptual Explanation

Efficiency = Time taken + Error rate

Satisfaction = Emotional response + Perceived control + Confidence

### Critical Analysis

Efficiency-focused systems often:

* Emphasize speed over clarity

* Reduce human interaction
* Ignore emotional reassurance

### HCI Example

Many online banking applications allow quick fund transfers (high efficiency). However:

* Frequent OTP authentication

* Fear of transaction failure
* Complex verification steps
reduce user satisfaction.

Users complete tasks quickly but feel stressed.

### Implications
Designers focusing only on performance metrics may overlook emotional design factors like
reassurance and transparency.

### Conclusion

Efficiency ensures task completion, but satisfaction determines long-term acceptance and
trust. Both must be balanced in HCI design.

---

# **Q13. Conflict Between Minimalism and Affordance**

### Introduction

Minimalist design reduces visual complexity, while affordance ensures users understand how
to interact with elements. These principles may conflict.

### Conceptual Explanation

* **Affordance**: Perceived action possibility.

* **Visibility**: How easily functions are discoverable.
* **Minimalism**: Reduction of visual clutter.

### Critical Analysis

Excessive minimalism may:

* Hide navigation options

* Remove visual cues
* Reduce discoverability
Overemphasis on affordance may:

* Create cluttered interface

* Increase cognitive load

### HCI Example

The hamburger menu icon hides multiple functions to maintain clean design. However, new
users may not discover features easily, reducing usability.

### Design Balance Strategy

1. Use progressive disclosure

2. Maintain clear signifiers
3. Prioritize primary functions

### Conclusion

Minimalism should simplify without obscuring functionality. Effective design balances

simplicity with discoverability.

---

# **Q14. Evaluation Approach for Real-World Deployment**

### Introduction

Systems may perform well in controlled usability testing but fail in real-world environments
due to contextual differences.
### Recommended Approach: Field Study / Longitudinal Evaluation**

### Conceptual Explanation

Field studies observe users in natural settings over extended periods. Longitudinal studies
capture behavioral changes over time.

### Why Effective

1. Captures environmental distractions

2. Identifies real workload conditions
3. Reveals long-term usability issues
4. Observes authentic behavior

### HCI Example

A mobile learning app tested in lab may perform well. However, in real use:

* Network issues
* Multitasking
* Fatigue
affect performance and satisfaction.

### Implications

Lab testing alone lacks ecological validity.

### Conclusion

Field and longitudinal evaluations provide realistic insights, preventing deployment failures
in practical environments.
---

# **Q15. Naive Bayes and Cognitive Load Dataset**

---

## (a) Modeling with Naive Bayes

### Introduction

Naive Bayes is a probabilistic classifier based on Bayes’ theorem and conditional

independence assumption.

### Modeling Process

For features M, E, T, K, N:

[
P(C|M,E,T,K,N) ∝ P(C)P(M|C)P(E|C)P(T|C)P(K|C)P(N|C)
]

Steps:

1. Compute class prior probabilities P(Low), P(High).

2. Compute conditional probabilities for each feature value given class.
3. Multiply probabilities.
4. Select class with higher posterior probability.

### Interpretation
Each interaction feature contributes independently to cognitive load prediction.

---

## (b) Violating Conditional Independence

Likely dependent features:

1. **Task Time (T) and Error Rate (E)**

Higher errors typically increase completion time.

2. **Keystroke Pause (K) and Task Time (T)**

Longer pauses directly increase total task duration.

These behavioral dependencies violate Naive Bayes assumption.

---

## (c) Efficient Navigation with High Errors**

This inconsistency suggests:

* Features may not fully represent cognitive state.

* Human behavior is multidimensional.
* Efficient navigation does not guarantee low cognitive load.

Feature-based models oversimplify complex psychological states.

---
## (d) Risk of Real-Time Adaptation Misclassification**

If high-load user misclassified as low-load:

* System remains complex.

* Cognitive overload increases.

If low-load user misclassified as high-load:

* Interface simplified unnecessarily.

* User may feel restricted.

### Impact

* Reduced trust
* Poor personalization
* User frustration

### Conclusion

While probabilistic classification enables adaptive HCI, misclassification risks must be

carefully managed to preserve user experience.

---