Mentora
Sign In
Official 50-Mark Examination BlueprintIncludes Step-Wise Marking Rubrics1-Click Printable PDF Format

ASSEB Higher Secondary AI Model Question Paper 2026

Practice authentic Higher Secondary (HS 1st & 2nd Year) Artificial Intelligence examination papers based on the official ASSEB / AHSEC 100-mark evaluation structure: 50 Marks Theory (3 Hours) + 50 Marks Practical (3 Hours). Includes model solutions, step-by-step examiner marking rubrics, and board viva questions.

Filter Sections:
ASSEB Official BlueprintSubject Code: 527 / 985 • W.E.F. 2025–26

Artificial Intelligence (HS 1st Year) — Model Question Paper 1

Duration: 3 Hours
Max Marks: 50
Section A
8 Marks
8 Qs × 1M
Section B
7 Marks
7 Qs × 1M
Section C
10 Marks
5 Qs × 2M
Section D
15 Marks
5 Qs × 3M
Section E
10 Marks
2 Qs × 5M
Q.1Section A• Unit 1: Introduction to AI
[1 Mark]

Which landmark test was proposed by Alan Turing in 1950 to evaluate if a computer possesses human-equivalent conversational intelligence?

(A)The Lovelace Machine Intelligence Index
(B)The Imitation Game (Turing Test)
(C)The Church-Turing Computational Thesis
(D)The Von Neumann Architecture Benchmark
Official Model Solution:
Option B: The Imitation Game (Turing Test)
Examiner Step-Wise Rubric:
  • 1 mark for selecting Option B: The Imitation Game (Turing Test)
Q.2Section A• Unit 2: AI Paradigms
[1 Mark]

In machine learning classification, which learning paradigm relies strictly on labeled dataset pairs containing both inputs (X) and ground truth targets (Y)?

(A)Unsupervised Learning
(B)Supervised Learning
(C)Reinforcement Learning without rewards
(D)Self-Organizing Clustering
Official Model Solution:
Option B: Supervised Learning
Examiner Step-Wise Rubric:
  • 1 mark for selecting Option B: Supervised Learning
Q.3Section A• Unit 3: Mathematics for AI
[1 Mark]

Which measure of central tendency is most robust and least affected when extreme outlier values appear in an asymmetric dataset?

(A)Arithmetic Mean
(B)Median
(C)Standard Deviation
(D)Variance
Official Model Solution:
Option B: Median
Examiner Step-Wise Rubric:
  • 1 mark for selecting Option B: Median
Q.4Section A• Unit 4: Programming for AI
[1 Mark]

In Python, which NumPy method is utilized to determine the dimensional structure (number of rows and columns) of an ndarray without modifying its elements?

(A)arr.length()
(B)arr.size()
(C)arr.shape
(D)arr.dim()
Official Model Solution:
Option C: arr.shape
Examiner Step-Wise Rubric:
  • 1 mark for selecting Option C: arr.shape
Q.5Section A• Unit 5: Data Collection & Exploration
[1 Mark]

When handling missing values in tabular datasets using Pandas, which function drops all rows containing at least one NaN entry?

(A)df.fillna()
(B)df.dropna()
(C)df.remove_nulls()
(D)df.isna().clear()
Official Model Solution:
Option B: df.dropna()
Examiner Step-Wise Rubric:
  • 1 mark for selecting Option B: df.dropna()
Q.6Section A• Unit 6: Storytelling with Data
[1 Mark]

Which visual plot is best suited for inspecting the correlation and joint distribution between two continuous numerical variables in Seaborn?

(A)Pie chart
(B)Scatter plot (or sns.scatterplot)
(C)Stacked bar chart
(D)Count plot
Official Model Solution:
Option B: Scatter plot (or sns.scatterplot)
Examiner Step-Wise Rubric:
  • 1 mark for selecting Option B: Scatter plot
Q.7Section A• Unit 7: AI Ethics & Bias
[1 Mark]

What type of algorithmic bias occurs when training dataset samples over-represent urban affluent communities and under-represent rural agricultural demographics?

(A)Algorithmic Quantization Bias
(B)Sampling / Representation Bias
(C)Hardware Precision Error
(D)Overfitting Bias
Official Model Solution:
Option B: Sampling / Representation Bias
Examiner Step-Wise Rubric:
  • 1 mark for selecting Option B: Sampling / Representation Bias
Q.8Section A• Unit 8: Employability Skills (ICT)
[1 Mark]

Which cyber-hygiene practice is essential to protect administrative accounts from credential stuffing attacks?

(A)Sharing passwords via unsecured plain SMS
(B)Enabling Multi-Factor Authentication (MFA / 2FA)
(C)Disabling system firewalls
(D)Using sequential birthdates as account PINs
Official Model Solution:
Option B: Enabling Multi-Factor Authentication (MFA / 2FA)
Examiner Step-Wise Rubric:
  • 1 mark for selecting Option B
Q.9Section B• Unit 1: Introduction to AI
[1 Mark]

Define Artificial Intelligence in one concise sentence according to the ASSEB curriculum framework.

Official Model Solution:
Artificial Intelligence is the branch of computer science dedicated to creating software and systems capable of simulating human cognitive capabilities—specifically learning, reasoning, pattern recognition, and autonomous problem solving.
Examiner Step-Wise Rubric:
  • 1 mark for defining AI as computational simulation of human cognitive capabilities (learning/reasoning/problem-solving).
Q.10Section B• Unit 2: AI Paradigms
[1 Mark]

State ONE key difference between Narrow AI (Weak AI) and Artificial General Intelligence (AGI).

Official Model Solution:
Narrow AI is engineered to execute one specific, bounded task (e.g. chess engines, facial recognition), whereas AGI is a theoretical system possessing cross-domain cognitive versatility equivalent to human intelligence.
Examiner Step-Wise Rubric:
  • 1 mark for contrasting single bounded domain vs versatile cross-domain human-level intelligence.
Q.11Section B• Unit 3: Mathematics for AI
[1 Mark]

Given a dataset of student quiz marks: [12, 14, 15, 18, 100]. Which central tendency indicator provides a more realistic typical score: Mean or Median?

Official Model Solution:
Median. The single extreme outlier (100) heavily skews the arithmetic mean upward, while the median (15) remains resilient.
Examiner Step-Wise Rubric:
  • 1 mark for identifying Median with reason (resilience against outlier 100).
Q.12Section B• Unit 4: Programming for AI
[1 Mark]

What does the method `df.info()` in Pandas output when called on a DataFrame?

Official Model Solution:
`df.info()` outputs a concise summary of the DataFrame including index range, column names, non-null value counts per column, data types, and total memory usage.
Examiner Step-Wise Rubric:
  • 1 mark for mentioning column names, non-null counts, datatypes, or memory summary.
Q.13Section B• Unit 5: Data Exploration
[1 Mark]

Why is feature scaling (e.g. Min-Max Normalization or Standardization) necessary before training distance-based algorithms such as K-Nearest Neighbors (KNN)?

Official Model Solution:
Feature scaling prevents features with large numeric scales (e.g. Annual Income in ₹) from disproportionately dominating Euclidean distance calculations over smaller scale features (e.g. Age in years).
Examiner Step-Wise Rubric:
  • 1 mark for explaining prevention of large-magnitude feature dominance in distance calculation.
Q.14Section B• Unit 7: AI Ethics
[1 Mark]

Define algorithmic accountability in the context of deploying automated decision systems.

Official Model Solution:
Algorithmic accountability is the principle that software creators and operating institutions must remain legally, ethically, and procedurally answerable for the real-world outcomes and biases of their AI models.
Examiner Step-Wise Rubric:
  • 1 mark for institutional/developer responsibility for algorithmic outcomes.
Q.15Section B• Part A: Green Skills
[1 Mark]

Name ONE eco-friendly practice in data center operations that supports the United Nations Sustainable Development Goals (SDGs).

Official Model Solution:
Powering server cooling systems with renewable solar/hydroelectric energy or adopting dynamic thermal load scheduling to reduce carbon footprint.
Examiner Step-Wise Rubric:
  • 1 mark for renewable energy, server power efficiency, or reduced e-waste.
Q.16Section C• Unit 2: AI Paradigms
[2 Marks]

Differentiate between Supervised Learning and Unsupervised Learning on the basis of (a) Training data format and (b) Primary objective.

Official Model Solution:
(a) Training Data: Supervised learning requires labeled pairs (inputs with target output ground-truth Y), whereas Unsupervised learning receives raw, unlabelled input data X. (b) Primary Objective: Supervised learning maps relationships to predict target labels for novel inputs, whereas Unsupervised learning discovers latent clusters, probability density, or patterns hidden within the data.
Examiner Step-Wise Rubric:
  • 1 mark for contrasting labeled vs unlabelled training data.
  • 1 mark for contrasting prediction/classification objective vs clustering/pattern discovery.
Q.17Section C• Unit 3: Mathematics for AI
[2 Marks]

Calculate the Euclidean distance between two 2D points: Point A = (2, 3) and Point B = (6, 6). Show the intermediate substitution step.

Official Model Solution:
Euclidean Distance formula: d = √[(x₂ - x₁)² + (y₂ - y₁)²] Substitution: d = √[(6 - 2)² + (6 - 3)²] = √[4² + 3²] = √[16 + 9] = √25 = 5 units.
Examiner Step-Wise Rubric:
  • 1 mark for stating formula and substituting values correctly: √[4² + 3²].
  • 1 mark for calculating final answer: 5 units.
Q.18Section C• Unit 4: Programming for AI
[2 Marks]

Write a short Python code snippet using NumPy to create a 3x3 matrix containing numbers from 1 to 9, and print its transpose.

Official Model Solution:
```python import numpy as np # Create 3x3 matrix from 1 to 9 matrix = np.arange(1, 10).reshape(3, 3) print('Original Matrix:\n', matrix) # Transpose using .T attribute or np.transpose() print('Transposed Matrix:\n', matrix.T) ```
Examiner Step-Wise Rubric:
  • 1 mark for correct creation/reshape of 3x3 array: np.arange(1, 10).reshape(3, 3).
  • 1 mark for transposing via .T or np.transpose() and printing.
Q.19Section C• Unit 5: Data Collection & Exploration
[2 Marks]

Explain the purpose of exploratory data analysis (EDA) before model building. List TWO typical anomalies detected during EDA.

Official Model Solution:
EDA allows data scientists to understand statistical distributions, detect missing values, and verify hypotheses before training. Two anomalies detected: (1) Outliers/sensor measurement glitches, and (2) High null-value rates or skewed imbalance across classes.
Examiner Step-Wise Rubric:
  • 1 mark for explaining the role of EDA in understanding data distribution and verifying assumptions.
  • 1 mark for identifying two anomalies (outliers, missing values, class skew, duplicate entries).
Q.20Section C• Unit 7: AI Ethics & Privacy
[2 Marks]

What is data anonymization? Explain why removing only candidate names is often insufficient to guarantee user privacy.

Official Model Solution:
Data anonymization is the process of removing or encrypting personal identifiers so data cannot be linked to a person. Merely removing names is insufficient because quasi-identifiers (such as ZIP code + Date of Birth + Gender) can be cross-linked with external public voter registries to re-identify individuals (known as linkage attacks).
Examiner Step-Wise Rubric:
  • 1 mark for definition of data anonymization.
  • 1 mark for explaining quasi-identifier linkage attacks (e.g. PIN code + DOB).
Q.21Section D• Unit 2: AI Paradigms
[3 Marks]

Explain Reinforcement Learning with the help of a neat conceptual diagram or cycle illustrating the interaction between the Agent, Environment, State, Action, and Reward.

Official Model Solution:
Reinforcement Learning (RL) is a goal-oriented paradigm where an autonomous Agent learns optimal decision policies through trial-and-error feedback from an Environment. Interaction Cycle: 1. State (S_t): Agent observes current environment situation. 2. Action (A_t): Agent chooses an action based on its policy. 3. Transition & Reward (R_t+1, S_t+1): The Environment updates to a new state and returns a scalar reward (+1 for victory, -1 for penalty). 4. Objective: Maximize cumulative expected future rewards over time.
Examiner Step-Wise Rubric:
  • 1 mark for defining Agent and Environment feedback dynamic.
  • 1 mark for describing State, Action, and Reward transitions.
  • 1 mark for stating the core objective: maximizing cumulative discounted return.
Q.22Section D• Unit 3: Mathematics for AI
[3 Marks]

A tea factory in Assam tests leaves using an automated optical sorter. The classification results on 100 test batches are summarized in the confusion matrix: - True Positives (TP) = 45 - False Positives (FP) = 5 - False Negatives (FN) = 10 - True Negatives (TN) = 40 Calculate: (a) Accuracy, (b) Precision, and (c) Recall.

Official Model Solution:
Total Samples = TP + TN + FP + FN = 45 + 40 + 5 + 10 = 100. (a) Accuracy = (TP + TN) / Total = (45 + 40) / 100 = 85 / 100 = 85% (0.85) (b) Precision = TP / (TP + FP) = 45 / (45 + 5) = 45 / 50 = 90% (0.90) (c) Recall = TP / (TP + FN) = 45 / (45 + 10) = 45 / 55 = 81.8% (0.818)
Examiner Step-Wise Rubric:
  • 1 mark for calculating Accuracy = 85%.
  • 1 mark for calculating Precision = 90%.
  • 1 mark for calculating Recall = 81.8%.
Q.23Section D• Unit 4: Programming for AI
[3 Marks]

Write a complete Python script using Pandas and Scikit-learn to: 1. Load a CSV dataset `'tea_data.csv'`. 2. Separate feature columns `'soil_ph'`, `'rainfall'` as `X` and target column `'yield_category'` as `y`. 3. Split data into 80% training and 20% testing sets using `train_test_split` with `random_state=42`.

Official Model Solution:
```python import pandas as pd from sklearn.model_selection import train_test_split # 1. Load CSV dataset df = pd.read_csv('tea_data.csv') # 2. Separate features X and target y X = df[['soil_ph', 'rainfall']] y = df['yield_category'] # 3. Train-test split (80-20) X_train, X_test, y_train, y_test = train_test_split( X, y, test_size=0.2, random_state=42 ) print(f'Training samples: {len(X_train)}, Test samples: {len(X_test)}') ```
Examiner Step-Wise Rubric:
  • 1 mark for correct pandas import and read_csv loading.
  • 1 mark for feature and target slicing: X = df[['soil_ph', 'rainfall']], y = df['yield_category'].
  • 1 mark for correct train_test_split syntax with test_size=0.2 and random_state=42.
Q.24Section D• Unit 6: Storytelling with Data
[3 Marks]

Compare the visualization strengths and use-cases of: (a) Histogram, (b) Box Plot, and (c) Heatmap.

Official Model Solution:
(a) Histogram: Displays frequency distribution of a single continuous variable across discrete bin ranges; ideal for checking normality or skewness. (b) Box Plot (Box & Whisker): Visualizes median, quartiles (Q1, Q3), interquartile range (IQR), and explicitly flags outliers beyond 1.5×IQR. (c) Heatmap: Represents 2D matrix correlations or cross-tabulations using chromatic color gradients; ideal for scanning multi-feature correlation matrices in EDA.
Examiner Step-Wise Rubric:
  • 1 mark for describing Histogram frequency distribution and bins.
  • 1 mark for describing Box Plot median/quartiles/outliers.
  • 1 mark for describing Heatmap color gradient matrix correlations.
Q.25Section D• Unit 7: AI Ethics & Assam Context
[3 Marks]

Discuss THREE ethical challenges when deploying automated facial recognition or AI monitoring systems in educational schools in Assam.

Official Model Solution:
1. Privacy and Consent: Minor students and rural guardians may lack informed technical awareness regarding biometric capture and biometric data storage policies. 2. Algorithmic Misidentification: Diverse regional indigenous facial profiles (e.g. Bodo, Mishing, Karbi) can suffer higher error rates if models were trained exclusively on Western or homogeneous facial databases. 3. Security Vulnerabilities: Biometric facial embeddings stored on unhardened school servers could be breached, creating permanent identity risks since human biometric traits cannot be reset like passwords.
Examiner Step-Wise Rubric:
  • 1 mark for explaining student privacy/informed consent.
  • 1 mark for explaining demographic representation and misidentification bias.
  • 1 mark for data security and permanent biometric leak hazards.
Q.26Section E• Unit 5 & 6: End-to-End AI Project Cycle
[5 Marks]

A high school team in Jorhat is developing an AI system to classify healthy vs diseased tea leaves (e.g. Red Rust / Blister Blight). Detail the FIVE steps of the AI Project Cycle for this initiative: 1. Problem Scoping 2. Data Acquisition 3. Data Exploration 4. Modelling 5. Evaluation

Official Model Solution:
1. Problem Scoping: Define the core agrarian challenge—early leaf disease diagnosis to prevent crop loss. Goal: Build an image/feature classifier that accurately flags infected tea bushes for smallholder farmers. 2. Data Acquisition: Collect high-resolution leaf photographs across healthy, red rust, and blister blight specimens across local tea gardens during morning and evening daylight conditions. 3. Data Exploration (EDA): Inspect dataset for blurriness, balance across disease classes, and label correctness. Resize images to standard dimensions (e.g. 224x224) and normalize pixel values. 4. Modelling: Train an initial Decision Tree / KNN / CNN classification model on 80% training data with supervised cross-entropy loss. 5. Evaluation: Evaluate model on unseen 20% test partition using Confusion Matrix, Precision, and Recall. Focus on minimizing False Negatives (diseased leaves classified as healthy) before deployment on farm mobile devices.
Examiner Step-Wise Rubric:
  • 1 mark for realistic problem scoping in tea agriculture.
  • 1 mark for data acquisition specifics (specimen collection across conditions).
  • 1 mark for data exploration & preprocessing (resizing, balancing).
  • 1 mark for model selection and training process.
  • 1 mark for evaluation metrics (precision, recall, false negative penalty).
Q.27Section E• Unit 4: Programming for AI
[5 Marks]

Consider the following Python code snippet using Scikit-learn: ```python from sklearn.neighbors import KNeighborsClassifier import numpy as np X = np.array([[1, 2], [2, 3], [3, 3], [6, 7], [7, 8], [8, 8]]) y = np.array([0, 0, 0, 1, 1, 1]) clf = KNeighborsClassifier(n_neighbors=3) clf.fit(X, y) new_sample = np.array([[2.5, 2.5]]) prediction = clf.predict(new_sample) ``` (a) What does `n_neighbors=3` signify? (b) Which class (`0` or `1`) will be predicted for `new_sample = [2.5, 2.5]`? Justify by calculating or comparing distances. (c) What happens if `n_neighbors` is set to an even number like `2` or `4` in binary classification?

Official Model Solution:
(a) `n_neighbors=3` specifies the parameter K—the algorithm will locate the 3 closest training points in feature space to cast majority votes for the query sample. (b) Prediction: Class `0`. Justification: Distance to points in Class 0: - to [2, 3]: √[(2.5-2)² + (2.5-3)²] = √[0.25 + 0.25] = √0.50 ≈ 0.707 - to [3, 3]: √[(2.5-3)² + (2.5-3)²] = √[0.25 + 0.25] = √0.50 ≈ 0.707 - to [1, 2]: √[(2.5-1)² + (2.5-2)²] = √[2.25 + 0.25] = √2.50 ≈ 1.581 Points in Class 1 are located around (6, 7), (7, 8), (8, 8) with distances > 5 units. All 3 nearest neighbors belong to Class 0, yielding unanimous prediction of Class 0. (c) When K is even in binary classification, a tie-vote (e.g. 1 vote for Class 0 and 1 vote for Class 1) can occur, forcing the implementation to fall back to arbitrary index priority or distance weighting.
Examiner Step-Wise Rubric:
  • 1 mark for explaining K=3 as voting threshold of nearest points.
  • 2 marks for calculating/comparing distance to Class 0 and predicting 0.
  • 2 marks for explaining tie-breaking ambiguity with even K in binary classification.

Need the official syllabus and unit-wise weightage?

View the official 100-mark ASSEB Artificial Intelligence syllabus with unit periods and textbook errata.

Official SyllabusPractice Board Mock

Official Examination FAQs & Evaluation Blueprint

What is the pass criteria for ASSEB AI (Subject 527)?

Students must score a minimum of 30% marks independently in the 50-mark Theory examination (15 Marks) and 30% marks in the 50-mark Practical examination (15 Marks), achieving an aggregate of at least 30 out of 100 overall.

Is Python coding tested in both Theory and Practical?

Yes. In Theory, Python is assessed through code tracing, syntax error debugging, and algorithm writing in Section C, D, and E. In Practical, students execute live Python scripts on lab workstations under external examiner supervision.

Are non-science students (Arts and Commerce) penalized in coding?

No. The ASSEB syllabus has NO high-level calculus or linear algebra prerequisites. All mathematics is taught intuitively using basic statistics (mean, median, Euclidean distance) and visual logic suitable for all three streams.

Can schools conduct the practical exam on Google Colab?

Yes. Schools with standard internet connectivity may use Google Colab with free CPU/T4 runtimes, while schools without constant internet can execute identical scripts offline using local Python 3.10 and VS Code.