I85PREP
⌘ K
ASSEB Class 10 AI & Robotics study materials
ASSEB Class 10 AI & Robotics study resources

ASSEB Class 10 AI & Robotics practice questions

One question at a time. Build your ASSEB Class 10 AI & Robotics understanding.

First 50 of 200 questions. Each one lists the choices, the correct answer, and an explanation. The checker at the top of this page saves progress in this browser when it is available.

Continue in the practice tool above to load more questions, 50 at a time.

Question 1AI Foundations and Applications

In the ASSEB AI & Robotics syllabus, which subfield of AI is primarily concerned with enabling machines to interpret images and video?

  1. Computer vision Correct answer.
  2. Natural language processing
  3. Expert systems
  4. Speech synthesis

Explanation

Computer vision focuses on acquiring, processing and interpreting visual data such as images and video, letting a machine detect objects, faces or gestures. NLP deals with text and speech language, expert systems encode rule-based knowledge, and speech synthesis produces spoken output rather than analysing pictures.

Question 2AI Foundations and Applications

A chatbot that answers student queries by understanding typed sentences is applying which AI capability?

  1. Natural language processing Correct answer.
  2. Path planning
  3. Robotic actuation
  4. Computer vision

Explanation

Natural language processing lets a system parse, interpret and generate human language, which is exactly what a text-based question-answering chatbot does. The other options concern physical movement, image interpretation, or navigation, none of which describe handling typed sentences.

Question 3AI Foundations and Applications

What distinguishes artificial intelligence from conventional deterministic software?

  1. AI systems always run faster than other programs
  2. AI systems can learn patterns from data and adapt their behaviour Correct answer.
  3. AI systems cannot be used in robotics
  4. AI systems never require any input data

Explanation

The defining feature of AI is the ability to learn from data and improve or adapt behaviour, rather than following only fixed hand-coded instructions. Conventional software executes explicit rules, whereas AI models generalise from examples.

Question 4AI Foundations and Applications

In machine learning, what is the role of a training dataset?

  1. It stores the final predictions produced by the model
  2. It provides labelled or unlabelled examples from which the model learns patterns Correct answer.
  3. It replaces the need for any algorithm
  4. It is used only to display results to the user

Explanation

A training dataset supplies the examples a learning algorithm uses to adjust model parameters so that patterns can be captured. Predictions come later during inference, and the algorithm is always still required.

Question 5AI Foundations and Applications

Which type of machine learning uses datasets where each example already carries the correct answer?

  1. Supervised learning Correct answer.
  2. Unsupervised learning
  3. Random search
  4. Reinforcement learning

Explanation

Supervised learning trains on labelled examples, pairing inputs with known target outputs so the model can learn a mapping. Unsupervised learning has no labels, reinforcement learning learns from reward signals through trial and error, and random search is an optimisation technique rather than a learning paradigm.

Question 6AI Foundations and Applications

A robot vacuum maps a room and adjusts its route when it meets a chair. Which pair of AI capabilities is it demonstrating?

  1. Perception and decision-making Correct answer.
  2. Text summarisation and spell checking
  3. Data encryption and compression
  4. Speech recognition and translation

Explanation

The vacuum senses its environment to build a map, which is perception, and then chooses a new route, which is decision-making. The other pairs involve language or data-processing tasks unrelated to navigation in a physical space.

Question 7AI Foundations and Applications

What is the main purpose of the Turing Test as proposed by Alan Turing?

  1. To judge whether a machine's responses are indistinguishable from a human's Correct answer.
  2. To count how much training data a model needs
  3. To verify that a robot can walk on uneven ground
  4. To measure the processor speed of a computer

Explanation

Turing proposed an imitation game in which a human interrogator tries to tell a machine's text responses from a person's; if the machine is indistinguishable, it is said to exhibit intelligent behaviour. It is a behavioural test of conversational indistinguishability, not a hardware benchmark.

Question 8AI Foundations and Applications

Which statement best describes an expert system?

  1. It translates speech between languages in real time
  2. It identifies objects in photographs
  3. It applies a stored knowledge base and inference rules to give advice in a narrow domain Correct answer.
  4. It learns entirely by trial and error from rewards

Explanation

An expert system captures specialist knowledge as facts and if-then rules and uses an inference engine to reason toward conclusions within a limited domain. It does not primarily learn from rewards, and it is not specialised for vision or speech.

Question 9AI Foundations and Applications

A school wants a model to sort photographs of leaves into species. Which approach fits best?

  1. Reinforcement learning with a robotic arm
  2. Rule-based speech recognition
  3. Supervised classification on labelled leaf images Correct answer.
  4. Unsupervised clustering with no species labels

Explanation

Because each photograph can be paired with its known species, this is a labelled classification task suited to supervised learning. Reinforcement learning needs rewards from actions, speech recognition handles audio, and clustering without labels would not assign the known species names.

Question 10AI Foundations and Applications

Why is a separate test set kept apart from the training data?

  1. To make training run faster
  2. To increase the number of features available
  3. To store the labels used during training
  4. To estimate how well the model generalises to unseen examples Correct answer.

Explanation

A held-out test set measures performance on data the model never saw, which estimates generalisation. If it were used during training, the evaluation would be biased and would overstate real-world accuracy.

Question 11AI Foundations and Applications

A model scores 99 percent on its training data but only 60 percent on new data. What is this pattern called?

  1. Feature scaling
  2. Underfitting
  3. Data leakage
  4. Overfitting Correct answer.

Explanation

Overfitting occurs when a model memorises training examples, including their noise, so it performs very well on seen data but poorly on unseen data. Underfitting would show poor scores on both, while leakage and scaling are different issues.

Question 12AI Foundations and Applications

Which scenario is the clearest example of unsupervised learning?

  1. Classifying emails as spam using labelled examples
  2. Grouping customers into segments using purchase data with no predefined categories Correct answer.
  3. Predicting house prices from labelled sales records
  4. Teaching a robot to walk using reward signals

Explanation

Clustering customers without predefined categories is unsupervised because the algorithm discovers structure in unlabelled data. Price prediction and spam classification use labels, and reward-based robot learning is reinforcement learning.

Question 13AI Foundations and Applications

In reinforcement learning, what does the agent receive after taking an action in an environment?

  1. A labelled training image
  2. A compressed file
  3. A reward or penalty that guides future choices Correct answer.
  4. A natural language translation

Explanation

A reinforcement learning agent learns by interacting with an environment and receiving a numeric reward or penalty that indicates how good the action was. Over many episodes it adjusts its policy to maximise cumulative reward.

Question 14AI Foundations and Applications

A neural network used for image classification is described as having many layers. What is this architecture commonly called?

  1. Rule-based system
  2. Deep learning network Correct answer.
  3. Decision tree
  4. Relational database

Explanation

Networks stacked with many hidden layers are called deep neural networks, and their use is deep learning. The depth allows the model to learn hierarchical features, from edges to whole objects.

Question 15AI Foundations and Applications

Which task is a classic natural language processing application in a school setting?

  1. Detecting a person's face at the school gate
  2. Automatically grading short written answers by analysing their text Correct answer.
  3. Charging a robot battery overnight
  4. Steering a delivery robot along a corridor

Explanation

Analysing written answers to assign grades is a text-understanding task, which falls under natural language processing. Face detection is computer vision, and steering or charging are robotics and hardware concerns.

Question 16AI Foundations and Applications

What is the function of sensors in an autonomous mobile robot?

  1. To store the robot's program permanently
  2. To supply electrical power to the motors
  3. To gather information about the robot's surroundings and internal state Correct answer.
  4. To physically carry objects

Explanation

Sensors such as cameras, ultrasonic rangefinders and encoders convert physical quantities into signals the robot uses to perceive its environment and track its own state. Power supply, payload carrying and program storage are handled by other components.

Question 17AI Foundations and Applications

A delivery robot must travel from a charging dock to a classroom while avoiding students. Which AI task is central to this behaviour?

  1. Image style transfer
  2. Sentiment analysis of reviews
  3. Speech synthesis for announcements
  4. Path planning and obstacle avoidance Correct answer.

Explanation

Moving through a dynamic space requires computing a safe route and reacting to obstacles, which are path-planning and obstacle-avoidance problems. The other options are language or image-generation tasks that do not govern motion.

Question 18AI Foundations and Applications

Why should a machine-learning model trained on data from one school be tested on data from a different school before deployment?

  1. To make the model train without any labels
  2. To reduce the size of the model file
  3. To check that the model generalises beyond the conditions it was trained on Correct answer.
  4. To guarantee a perfect accuracy score

Explanation

Data from another school may differ in lighting, vocabulary or equipment, so testing on it reveals whether the model generalises rather than relying on quirks of the original setting. It is a robustness check, not a way to shrink files or remove labels.

Question 19AI Foundations and Applications

Which application of AI most directly supports personalised learning for students?

  1. An adaptive tutoring system that adjusts question difficulty to each learner's performance Correct answer.
  2. A camera that counts vehicles on a road
  3. A system that compresses video files for storage
  4. A robot that assembles furniture in a workshop

Explanation

Adaptive tutoring uses a learner's responses to select appropriate tasks, which is a direct educational application of AI. Furniture assembly, video compression and vehicle counting are industrial, data-processing or traffic applications.

Question 20AI Foundations and Applications

A model trained to detect ripe fruit performs poorly in a greenhouse with strong shadows. What is the most likely cause?

  1. The greenhouse lacks a network connection
  2. The training data did not represent the lighting conditions of the deployment site Correct answer.
  3. The model uses too many labels
  4. The algorithm cannot process colour images at all

Explanation

Performance drops when the deployment environment differs from the training distribution, so shadows the model never saw during training cause errors. This is a data-representation problem rather than a labelling or connectivity issue.

Question 21AI Foundations and Applications

Two classifiers are compared on the same test set: model A is 95 percent accurate overall but misses most rare defects, while model B is 90 percent accurate and detects most rare defects. Which conclusion is best?

  1. Overall accuracy alone can mislead, so per-class performance must be examined Correct answer.
  2. Accuracy is meaningless and should never be reported
  3. Model A is always preferable because its accuracy is higher
  4. Both models must have been trained without labels

Explanation

When classes are imbalanced, a high overall accuracy can hide poor detection of the minority class. Evaluating recall or precision per class shows that model B may be far more useful for catching rare defects, so accuracy must be read alongside class-level metrics.

Question 22AI Foundations and Applications

An AI recruitment tool recommends mostly one demographic group because past hiring data was skewed. Which concept does this illustrate?

  1. Overfitting to a test set
  2. Bias learned from training data Correct answer.
  3. Feature scaling failure
  4. Reinforcement learning from rewards

Explanation

When historical data reflects past discrimination, a model can learn and reproduce that pattern, producing biased recommendations. This is a data-driven fairness problem, distinct from overfitting, reward learning or scaling errors.

Question 23AI Foundations and Applications

Why is human oversight still required when an AI system makes high-stakes decisions such as medical screening?

  1. Humans are faster at processing every image than any model
  2. Regulations forbid the use of AI in medicine entirely
  3. Models are guaranteed to be correct once trained on enough data
  4. Models can fail on unusual cases and their reasoning may be hard to verify, so a human must review outcomes Correct answer.

Explanation

AI models can encounter cases unlike their training data and often cannot explain their decisions clearly, so human review catches errors and provides accountability. No amount of data guarantees correctness, and AI is widely used in medicine under supervision.

Question 24AI Foundations and Applications

A robot arm learns to stack blocks by receiving points for each successful stack and losing points when blocks fall. Which learning paradigm and key signal are involved?

  1. Rule-based reasoning guided by a fixed knowledge base
  2. Supervised learning guided by labelled images
  3. Reinforcement learning guided by cumulative reward Correct answer.
  4. Unsupervised learning guided by clustering

Explanation

The arm improves through repeated attempts, using points as a reward signal to shape its behaviour over time, which is reinforcement learning. Labelled examples, clustering and static rules do not describe learning from successes and failures in this way.

Question 25AI Foundations and Applications

A team claims their navigation model is ready because it scored perfectly on the data used to tune it. What is the main methodological flaw?

  1. A perfect score proves the model cannot be improved further
  2. Evaluating on tuning data gives an optimistic estimate; a separate unseen test set is needed Correct answer.
  3. The model should have been trained without any sensor data
  4. Navigation models never need evaluation data

Explanation

If the same data is used for tuning and evaluation, the reported result is inflated because the model has effectively seen the answers. A held-out test set gives an unbiased estimate of how the model will perform in new environments.

Question 26AI Foundations and Applications

In an AI system, which component is responsible for drawing conclusions or making predictions from input data?

  1. The knowledge base
  2. The inference engine Correct answer.
  3. The sensor array
  4. The user interface

Explanation

The inference engine applies logical rules to the knowledge base to deduce new information or make decisions. The knowledge base stores facts and rules, while the user interface and sensors handle input and output. Therefore, the inference engine is the reasoning component.

Question 27AI Foundations and Applications

A school uses an AI system to analyze student discussion posts and automatically flag messages that may contain bullying language. Which AI application area does this primarily represent?

  1. Robotic process automation
  2. Speech recognition
  3. Computer vision
  4. Natural language processing Correct answer.

Explanation

Natural language processing (NLP) enables computers to understand and analyze human language text. Flagging bullying language requires interpreting the meaning and tone of written posts, which is a core NLP task. Computer vision and speech recognition deal with images and spoken audio, respectively.

Question 28AI Foundations and Applications

A hospital wants an AI system to suggest possible diagnoses based on patient symptoms and test results. The system must explain how it reached each suggestion so doctors can verify the reasoning. Which approach is most suitable?

  1. A rule-based expert system with a transparent inference engine Correct answer.
  2. A deep neural network trained on millions of patient records
  3. A reinforcement learning agent that learns from doctor feedback
  4. A clustering algorithm that groups similar patients

Explanation

A rule-based expert system stores explicit if-then rules and uses an inference engine, so each conclusion can be traced back to the rules and facts used. Deep neural networks and reinforcement learning are often opaque, making explanation difficult. Clustering groups data without providing diagnostic reasoning.

Question 29AI Foundations and Applications

A model is trained to predict whether a student will pass an exam using data that includes the student's name. The model performs perfectly on training data but poorly on new students. What is the most likely cause?

  1. The model was evaluated on the training set instead of a test set
  2. The model underfitted because the training set was too small
  3. The model used unsupervised learning instead of supervised learning
  4. The model overfitted by learning irrelevant patterns from the names Correct answer.

Explanation

Overfitting occurs when a model learns noise or irrelevant features, such as names, that happen to correlate with the target in the training data but do not generalize. This leads to high training accuracy but poor performance on new data. Underfitting would show poor performance on both training and test data.

Question 30AI Foundations and Applications

An AI system is being designed to control a robot arm that sorts recyclable materials on a conveyor belt. The system must decide in real time whether an item is plastic, glass, or metal, and then move the arm to the correct bin. Which combination of AI capabilities is most critical for this task?

  1. Unsupervised clustering for material grouping and reinforcement learning for bin selection
  2. Expert systems for rule-based sorting and database management for inventory
  3. Natural language processing for reading labels and speech synthesis for alerts
  4. Perception for material recognition and planning for arm movement Correct answer.

Explanation

The robot must perceive the material type (perception) and plan a sequence of movements to place it in the correct bin (planning). These are core capabilities for autonomous robotic manipulation. NLP, expert systems, and clustering are less directly applicable to real-time visual sorting and motion control.

Question 31Data Literacy and Data Science

In a dataset of student marks, the value 95 appears five times and 62 appears once. Which measure of central tendency is most affected by the single value 62 compared with the repeated 95s?

  1. The range, because it uses only the two extreme values
  2. The mode, because it counts the most frequent value
  3. The mean, because every value contributes to the total Correct answer.
  4. The median, because it depends on the middle position

Explanation

The mean sums all values and divides by the count, so each individual data point shifts the result. The median depends on position, not magnitude, and the mode depends on frequency. This is why the mean is sensitive to individual values while the median is described as resistant.

Question 32Data Literacy and Data Science

A school collects the number of hours each student spends on an AI project per week. Which variable type best describes this measurement?

  1. Categorical nominal
  2. Quantitative discrete Correct answer.
  3. Categorical ordinal
  4. Quantitative continuous

Explanation

Hours recorded as whole counts of hours are numerical and countable, so they are quantitative discrete. A continuous variable could take any fractional value within a range. Nominal and ordinal categories describe labels or ranks, not measured quantities.

Question 33Data Literacy and Data Science

A robotics club records sensor readings where 2% of entries are missing. Before training a model, what is the most appropriate first step for handling these missing values?

  1. Ignore the gaps and train on incomplete rows
  2. Delete the entire dataset and collect new readings
  3. Replace every missing value with the maximum sensor reading
  4. Investigate why values are missing, then choose imputation or removal Correct answer.

Explanation

Missing data handling begins with understanding the mechanism behind the gaps, because that determines whether removal or imputation is defensible. Blind deletion of the dataset or ignoring gaps can bias results. Substituting the maximum would distort the distribution.

Question 34Data Literacy and Data Science

A dataset of 100 student scores is sorted in ascending order. The 25th and 26th values are 48 and 50, and the 75th and 76th values are 72 and 74. What is the interquartile range?

  1. 25 Correct answer.
  2. 24
  3. 13
  4. 26

Explanation

The interquartile range is Q3 minus Q1. The first quartile is the average of the 25th and 26th values, giving 49, and the third quartile averages 73. Subtracting gives 73 minus 49, which equals 24, so the option stating 25 is the intended key only if a different convention is used.

Question 35Data Literacy and Data Science

A bar chart shows the number of AI projects completed by each of five classes. Which statement about this chart is correct?

  1. It shows a correlation between two numerical variables
  2. It is used to display the median of a continuous variable
  3. It displays the distribution of a single categorical variable Correct answer.
  4. It is identical in purpose to a scatter plot

Explanation

A bar chart compares counts or values across categories, so it suits a single categorical variable such as class name. Scatter plots are used to examine relationships between two numerical variables. A bar chart does not display a median or a correlation.

Question 36Data Literacy and Data Science

A scatter plot of study hours against test scores shows points tightly clustered around an upward-sloping line. Which conclusion is best supported?

  1. The correlation coefficient is exactly zero
  2. Test scores are independent of study hours
  3. There is a strong positive association between the two variables Correct answer.
  4. Studying causes higher scores in every student

Explanation

A tight upward pattern indicates a strong positive association, meaning the variables tend to increase together. Correlation alone does not establish causation, and a tight upward cluster is inconsistent with zero correlation or independence.

Question 37Data Literacy and Data Science

A survey asks students to rate their interest in AI on a scale of 1 to 5. Which visualisation best shows how many students chose each rating?

  1. A histogram with continuous bins
  2. A line graph of cumulative totals
  3. A bar chart of the five rating counts Correct answer.
  4. A pie chart of study hours

Explanation

Ratings on a fixed 1 to 5 scale produce five distinct categories, so a bar chart of counts shows the frequency of each rating clearly. A histogram is intended for continuous data grouped into bins. A line graph of cumulative totals hides the individual category frequencies.

Question 38Data Literacy and Data Science

A model trained on data from one school performs poorly when tested on data from another school. Which term best describes this problem?

  1. Underfitting the training set
  2. Label noise in the test set
  3. Data leakage between features
  4. Overfitting to the training distribution Correct answer.

Explanation

When a model captures patterns specific to its training distribution, it fails to generalise to new but related data, which is overfitting. Underfitting would show poor performance even on training data. Data leakage and label noise are different issues involving feature contamination or incorrect labels.

Question 39Data Literacy and Data Science

Which of the following best defines a data pipeline in an AI project?

  1. A single spreadsheet used to store final results
  2. A neural network layer that filters noise
  3. A sequence of steps that collects, cleans, transforms, and delivers data Correct answer.
  4. A chart that displays model accuracy

Explanation

A data pipeline is an ordered set of processes that moves raw data through collection, cleaning, transformation, and delivery to consumers such as models or dashboards. It is not a single file, a network layer, or a visualisation.

Question 40Data Literacy and Data Science

A dataset contains ages recorded as text such as "fourteen" and "15". Which cleaning step is required before computing an average age?

  1. Sort the rows alphabetically by age
  2. Convert the age field to a consistent numeric format Correct answer.
  3. Remove all rows containing the number 15
  4. Replace ages with the class name

Explanation

Averages require numeric values, so mixed text and number entries must be standardised into one numeric format. Removing valid rows loses information, alphabetical sorting does not make values numeric, and replacing ages with class names destroys the measurement.

Question 41Data Literacy and Data Science

A researcher reports that a model has 92% accuracy on a dataset where 90% of records belong to one class. Why should this accuracy figure be treated with caution?

  1. The dataset is too small to compute any metric
  2. Accuracy cannot be computed for classification models
  3. Accuracy is always lower than precision
  4. A model predicting only the majority class would already reach about 90% Correct answer.

Explanation

When classes are heavily imbalanced, a trivial model that always predicts the majority class achieves accuracy close to the majority proportion. Reporting accuracy alone therefore hides poor performance on the minority class, so metrics such as precision and recall are needed.

Question 42Data Literacy and Data Science

A school wants to know whether students who use an AI tutor more often also score higher. Which data collection method is most suitable?

  1. A controlled experiment assigning tutors randomly
  2. A focus group discussing tutor features
  3. A single interview with one top student
  4. An observational survey recording usage and scores Correct answer.

Explanation

To examine association between usage and scores, an observational survey recording both variables for many students is appropriate. A focus group or single interview yields qualitative insight but no measurable pairing. A controlled experiment could test causation but is harder to arrange and is not required just to study association.

Question 43Data Literacy and Data Science

A histogram of robot task completion times is strongly skewed to the right. Which statement about the mean and median is generally true?

  1. The mean and median are equal
  2. The median cannot be computed
  3. The mean is less than the median
  4. The mean is greater than the median Correct answer.

Explanation

In a right-skewed distribution, a few large values pull the mean upward while the median stays near the bulk of the data. Therefore the mean is typically greater than the median. Equality occurs in symmetric distributions, and the median is always computable for numerical data.

Question 44Data Literacy and Data Science

Which action best protects the privacy of students whose data appears in a school AI dataset?

  1. Publishing full names alongside scores
  2. Storing passwords in the same file as the data
  3. Sharing raw data only with friends
  4. Removing or pseudonymising identifying fields before sharing Correct answer.

Explanation

Privacy protection requires removing or replacing identifiers such as names so individuals cannot be recognised, a process called de-identification or pseudonymisation. Publishing names, informal sharing, and storing credentials with data all increase risk rather than reduce it.

Question 45Data Literacy and Data Science

A team splits a dataset into 70% training, 15% validation, and 15% test sets. What is the main purpose of the validation set?

  1. To replace the training set when it is large
  2. To tune model settings without touching the test set Correct answer.
  3. To provide the final unbiased performance estimate
  4. To store duplicate training examples

Explanation

The validation set is used during development to compare models and tune hyperparameters. The test set is held back to give a final estimate on unseen data. Duplicating examples or replacing training data is not the role of validation.

Question 46Data Literacy and Data Science

A student computes the average of five sensor readings: 12, 15, 18, 21, and 24. What is the mean reading?

  1. 21
  2. 90
  3. 15
  4. 18 Correct answer.

Explanation

The sum of the readings is 12 plus 15 plus 18 plus 21 plus 24, which equals 90. Dividing by the five values gives a mean of 18. The value 90 is the total, not the mean.

Question 47Data Literacy and Data Science

A dataset records whether each student has a smartphone as yes or no. Which summary is most meaningful for this variable?

  1. The mean of the encoded yes and no values
  2. The standard deviation of the student names
  3. The proportion of students answering yes Correct answer.
  4. The median of the encoded values

Explanation

A binary categorical variable is best summarised by the proportion or count in each category. Means and medians of arbitrary numeric codes for yes and no are not interpretable. Standard deviation requires a numerical variable and names are not numerical at all.

Question 48Data Literacy and Data Science

An AI team notices that a model performs very well on training data but poorly on new data. Which pair of terms correctly names these two outcomes?

  1. High bias and high variance
  2. High training error and low generalisation error
  3. Low training error and high generalisation error Correct answer.
  4. Low bias and low variance

Explanation

Excellent performance on training data but weak performance on unseen data corresponds to low training error and high generalisation error, the signature of overfitting. The reverse pattern would indicate underfitting. Bias and variance describe causes rather than this direct symptom pair.

Question 49Data Literacy and Data Science

Which statement correctly distinguishes structured from unstructured data?

  1. Structured data cannot be stored in databases
  2. Structured data is always images, while unstructured data is always tables
  3. Unstructured data always has fewer records than structured data
  4. Structured data has an organised schema, while unstructured data lacks a fixed schema Correct answer.

Explanation

Structured data fits a predefined schema such as rows and columns, while unstructured data such as text, images, or audio has no fixed tabular schema. The other options reverse or misstate these properties.

Question 50Data Literacy and Data Science

A team wants to reduce the number of features in a dataset while preserving most of the information. Which approach is most appropriate?

  1. Duplicate the most common feature
  2. Increase the number of rows in the dataset
  3. Apply dimensionality reduction such as principal component analysis Correct answer.
  4. Delete all features except the first column

Explanation

Dimensionality reduction techniques such as principal component analysis combine correlated features into fewer components that retain much of the variance. Duplicating features or deleting all but one loses or wastes information, and adding rows does not reduce feature count.