The Three GATE DA Papers at a Glance
Every GATE DA paper so far has had 65 questions worth 100 marks in 3 hours. That breaks down into 10 General Aptitude questions worth 15 marks and 55 subject questions worth 85 marks (GATE 2027 Question Paper Pattern). Questions carry 1 or 2 marks and come in three types: multiple choice (MCQ), multiple select (MSQ) and numerical answer (NAT). Only MCQs have negative marking.
The official papers and answer keys are free on the GATE 2027 website, run by IIT Madras:
| Year | Organising institute | Official question paper | Official answer key |
|---|---|---|---|
| 2024 | IISc Bengaluru | GATE DA 2024 paper (PDF) | Answer key (PDF) |
| 2025 | IIT Roorkee | GATE DA 2025 paper (PDF) | Answer key (PDF) |
| 2026 | IIT Guwahati | GATE DA 2026 paper (PDF) | Answer key (PDF) |
All six files are linked from the official GATE 2027 download page. If you prefer to practise online, Success Tracker's GATE DA PYQ archive has every question from the three papers, sorted by year and by subject, with solutions.
Subject-Wise Marks, 2024 to 2026
The official pattern does not split the 85 subject marks between DA subjects. The table below is The ML Hub's own count: we assigned every question in the three official papers to one subject and added up its marks.
| Subject | 2024 | 2025 | 2026 | Average | Trend |
|---|---|---|---|---|---|
| Probability & Statistics | 15 | 19 | 19 | 17.7 | Highest-scoring subject, steady |
| Programming, Data Structures & Algorithms | 20 | 14 | 14 | 16.0 | Highest in 2024, steady since |
| Machine Learning | 14 | 14 | 13 | 13.7 | Steady, 13–14 marks |
| DBMS & Warehousing | 7 | 11 | 18 | 12.0 | Rising every year |
| Linear Algebra | 11 | 12 | 8 | 10.3 | Fell in 2026 |
| Artificial Intelligence | 10 | 6 | 9 | 8.3 | Steady, 6–10 marks |
| Calculus & Optimization | 8 | 9 | 4 | 7.0 | Fell sharply in 2026 |
| General Aptitude | 15 | 15 | 15 | 15.0 | Fixed by the pattern |
| Total | 100 | 100 | 100 | 100 |
This is the same count we use in our 3-month GATE DA plan. It is not an official split. A few questions touch two subjects (for example, a z-score question that is both statistics and data transformation), and we assigned each to the subject it mainly tests. For example, we count two 2025 questions under Machine Learning: Q48 on the properties of the ReLU activation (which also tests continuity and differentiability) and Q60 on the variance along the top principal component (which also uses covariance eigenvalues). Counted under Calculus and Linear Algebra instead, Machine Learning would show 10 marks in 2025. Three papers is a small sample, so read the trends as signals, not rules.
What the Totals Tell You
- Maths is over a third of the paper. Probability & Statistics, Linear Algebra and Calculus together carried 34 marks in 2024, 40 in 2025 and 31 in 2026, an average of about 35. If your maths is weak, fix that first.
- Programming and DBMS together are about a quarter. They carried 27, 25 and 32 marks. These are the most "CS-like" parts of the paper, and students from a CS background have an edge here.
- ML and AI are smaller than the paper's name suggests. Together they carried 24, 20 and 22 marks, about a fifth of the paper. The ML questions test fundamentals (regression, classifiers, clustering, PCA), not the latest models.
- Every subject has carried at least 4 marks every year. No subject is safe to skip.
Question Types: Where the Numerical and Multi-Select Questions Are
How a subject is asked matters as much as how many marks it carries. MCQs have negative marking. MSQs have no negative marking but also no partial credit, so you must pick every correct option. NAT questions need an exact number, so you can't guess from options (official marking rules).
The table below counts question types by subject across all three papers, as tagged on Success Tracker. A few questions carry two subject tags, so the totals (206) are slightly more than the 195 questions in the papers:
| Subject | Tagged questions | 1-mark / 2-mark | MCQ | MSQ | NAT |
|---|---|---|---|---|---|
| Probability & Statistics | 40 | 15 / 25 | 21 | 4 | 15 |
| Programming, DS & Algorithms | 32 | 15 / 17 | 16 | 9 | 7 |
| General Aptitude | 29 | 16 / 13 | 28 | 0 | 1 |
| Machine Learning | 28 | 13 / 15 | 13 | 7 | 8 |
| DBMS & Warehousing | 23 | 10 / 13 | 10 | 6 | 7 |
| Linear Algebra | 21 | 9 / 12 | 9 | 9 | 3 |
| Artificial Intelligence | 18 | 11 / 7 | 10 | 6 | 2 |
| Calculus & Optimization | 15 | 6 / 9 | 5 | 6 | 4 |
| All subjects | 206 | 95 / 111 | 112 | 47 | 47 |
- Probability & Statistics is the most numerical subject. 15 of its 40 tagged questions were NAT, and 25 were 2-mark questions. Practise working answers to the final number, not just the method.
- Linear Algebra leans on multi-select. 9 of 21 questions were MSQ, typically "which of these statements about this matrix are true". You need to be sure of every property, because one wrong pick loses the whole question.
- Calculus is MSQ-heavy too. 6 of its 15 questions were MSQ, usually statements about maxima, minima and continuity.
- General Aptitude is almost all MCQ. Careless guessing there costs negative marks.
Subject-by-Subject Topic Trends
Below, we grouped each subject's questions by topic. These groupings are our own classification of the questions on Success Tracker, so the counts are approximate. They are shown as total across the three papers, with the split by year in brackets (2024 / 2025 / 2026).
Probability & Statistics (40 questions)
| Topic | Questions (2024 / 2025 / 2026) | Typical question |
|---|---|---|
| Random variables and distributions | 16 (3 / 7 / 6) | Exponential distribution: find a mean or conditional probability from a given tail probability. Normal pdf and CDF statements. |
| Expectation, variance, covariance | 10 (4 / 4 / 2) | Covariance of two variables built from coin tosses. Law of total expectation, E[E[X|Y]]. |
| Counting | 8 (3 / 1 / 4) | Number of non-negative integer solutions to an equation. Counting numbers divisible by 3 from a digit set. |
| Conditional probability and Bayes | 3 (1 / 1 / 1) | A medical test with false positives. Balls drawn from one of three boxes. |
| Sampling and inference | 3 (1 / 1 / 1) | How a sample mean changes when a point is added. Statistics of an i.i.d. normal sample. |
Takeaway: random variables and their distributions are the single most-asked topic in the whole paper. Know the mean, variance and memoryless property of the exponential distribution, and be fast with the normal table. Hypothesis tests (z, t, chi-squared) are on the syllabus but have been asked rarely so far. Cover them, but after distributions.
Programming, Data Structures & Algorithms (32 questions)
| Topic | Questions (2024 / 2025 / 2026) | Typical question |
|---|---|---|
| Code tracing and recursion | 9 (3 / 2 / 4) | Trace a recursive Python function and give the printed value. |
| Sorting and searching | 8 (3 / 2 / 3) | Quicksort with the last element as pivot. Worst-case comparisons in binary search. Insertion sort swaps. |
| Graphs and hashing | 7 (3 / 3 / 1) | BFS and DFS orders and discovery times. Linear probing with a given hash function. |
| Stacks, queues, lists | 4 (2 / 2 / 0) | Deque operations. A stack-based pseudocode output. |
| Trees | 3 (2 / 0 / 1) | Rebuild a tree from pre-order and in-order traversals. Relations between leaves and internal nodes. |
Takeaway: the most common question type is "what does this code print". Practise tracing Python by hand, especially recursion. Know quicksort, mergesort and the simple sorts step by step, not just their complexity.
Machine Learning (28 questions)
| Topic | Questions (2024 / 2025 / 2026) | Typical question |
|---|---|---|
| Classification (k-NN, SVM, Naive Bayes, logistic) | 9 (5 / 3 / 1) | k-NN on five labelled points. A linear SVM on 2-D data. Parameter count for a Naive Bayes-style classifier. |
| Clustering and dimensionality reduction | 7 (2 / 3 / 2) | Single-linkage or agglomerative clustering from a distance table. PCA dimensions and variance. |
| Neural networks | about 4 | Output of a small ReLU network. Parameter count of an MLP. |
| Regularisation and validation | 3 (0 / 0 / 3) | Ridge regression statements. Choosing a cross-validation setup. |
| Regression | 2 (0 / 1 / 1) | Least-squares fit of y = wx to three points, and its gradient. |
Takeaway: GATE ML questions are small calculations done by hand: run k-NN, cluster four points, count parameters. Practise them with pen and paper. All three regularisation and validation questions came in 2026, which may be a new trend. We found no question mainly about decision trees, even though they are on the syllabus, so don't assume they will stay absent.
DBMS & Warehousing (23 questions)
| Topic | Questions (2024 / 2025 / 2026) | Typical question |
|---|---|---|
| SQL and relational algebra | 6 (1 / 2 / 3) | Output of an SQL query with aggregates or joins. A relational algebra expression on two tables. |
| ER model and constraints | 5 (1 / 2 / 2) | Mapping an ER diagram to tables. Foreign-key constraints. |
| Indexing, B+ trees, tuple calculus | about 4 | Maximum keys or pointers in a B+ tree node. A tuple relational calculus query. |
| Functional dependencies and normal forms | 4 (1 / 2 / 1) | Candidate keys from a set of FDs. Dependency-preserving decomposition. |
| Data warehousing and OLAP | 3 (0 / 1 / 2) | Number of cuboids from concept hierarchies. Operations on a sales data cube. |
Takeaway: DBMS & Warehousing has grown every year, from 7 to 11 to 18 marks. Warehousing questions only started appearing in 2025 and 2026. Learn data cubes and cuboid counting, which take little time.
Linear Algebra (21 questions)
| Topic | Questions (2024 / 2025 / 2026) | Typical question |
|---|---|---|
| Rank, nullity, subspaces | 7 (2 / 4 / 1) | Rank of I + xxT. Which sets are subspaces. Properties of idempotent and nilpotent matrices. |
| Eigenvalues and eigenvectors | 4 (3 / 0 / 1) | Eigenvalues of a rotation block. A determinant from known eigenvalues. |
| Orthogonality and projections | 4 (0 / 2 / 2) | Properties of the centering matrix I − (1/n)11T. |
| Systems of equations and elimination | 3 (1 / 1 / 1) | When a system is consistent. Cost of Gaussian elimination. |
| SVD and low rank | 3 (1 / 2 / 0) | Singular values of a matrix built from one vector. Sums of rank-one projections. |
Takeaway: GATE DA linear algebra is about matrix properties, not long computations. Learn what rank, projection, idempotent and orthogonal matrices imply, because most questions are true-or-false statements in MSQ form.
Artificial Intelligence (18 questions)
| Topic | Questions (2024 / 2025 / 2026) | Typical question |
|---|---|---|
| Propositional and first-order logic | 7 (2 / 1 / 4) | Logical equivalence of English statements. Which first-order statements are valid. |
| Adversarial search | 3 (1 / 1 / 1) | Minimax value or alpha-beta pruning on a game tree. |
| Bayesian networks and inference | 3 (2 / 1 / 0) | Probabilities and independence in a small Bayesian network. |
| Uninformed and informed search | 4 (2 / 1 / 1) | A* with a given heuristic. Combining admissible heuristics. DFS order. |
Takeaway: logic is the biggest AI topic and was 4 questions in 2026 alone. Minimax appeared every year.
Calculus & Optimization (15 questions)
| Topic | Questions (2024 / 2025 / 2026) | Typical question |
|---|---|---|
| Maxima, minima and derivatives | 8 (3 / 4 / 1) | Extrema of a cubic on an interval. Derivative of the sigmoid. |
| Limits, continuity, differentiability | 6 (2 / 3 / 1) | Evaluate a limit with logs or square roots. |
Takeaway: the whole subject reduces to two topics in past papers. It dropped to 4 marks in 2026, but it was 11 in 2025, so give it a few focused days.
General Aptitude (29 questions)
Quantitative reasoning was the largest group (10 questions), followed by vocabulary and analogies (6), spatial and visual puzzles such as paper folding and figure counting (5) and data interpretation from charts (4). It is worth 15 marks in every paper. Our General Aptitude guide covers how to prepare.
How to Use This Analysis
- Set your study order by marks. Start with Probability & Statistics and Programming, then ML and DBMS. Our 90-day GATE DA plan turns this table into a dated timetable.
- Solve every past question topic-wise as you learn. There are only 195, so solve them all. Use Success Tracker's subject pages for Probability & Statistics, Programming & DSA, Machine Learning, DBMS & Warehousing, Linear Algebra, AI, Calculus and General Aptitude.
- Then take each paper as a timed test. Use the official PDFs above, set a 3-hour timer, and mark with the official key.
- Practise by question type. Do NAT practice for probability, and true-or-false statement drills for linear algebra and calculus.
- Check the 2027 syllabus. GATE 2027 has a revised syllabus. Our GATE DA 2027 syllabus changes post compares it line by line with 2026, so you know which past questions still apply.
The ML Hub's GATE DA test series follows the same subject mix as the official papers.
View the GATE DA Test Series → | Explore the GATE DA Course →
FAQs
How many GATE DA previous year papers are there?
Three: 2024, 2025 and 2026. GATE DA was first held in 2024, and each year had one paper of 65 questions, so there are 195 official questions in total. All three papers and answer keys are on the official GATE 2027 download page.
Which subject has the highest weightage in GATE DA?
By our count of the three official papers, Probability & Statistics, with 15, 19 and 19 marks (17.7 on average). Programming, Data Structures & Algorithms is next at 16.0. There is no official subject-wise split.
Is GATE DA more maths or more ML?
More maths. Probability, Linear Algebra and Calculus together averaged about 35 marks a paper in our count, while Machine Learning and AI together averaged about 22.
Do GATE DA questions repeat?
Topics repeat, even when the exact questions don't. Exponential and normal distributions, recursive code tracing, sorting, clustering by hand, SQL output and logic equivalence have appeared in every paper.
Are GATE CS questions useful for GATE DA?
For the overlapping topics, yes. Programming and data structures, DBMS, linear algebra and probability overlap with the CS syllabus, so CS past questions on those topics are good extra practice. Note that DA programming is in Python, while CS uses C.
Related Reading
- GATE DA 2027 Preparation in 3 Months: the 90-day timetable built on this analysis
- GATE DA Syllabus 2027: every line of the official syllabus explained
- GATE DA Marks vs Score: how raw marks become a GATE score
- GATE DA Books 2027: one book per subject
Sources
- GATE 2027 Download page: official GATE DA 2024, 2025 and 2026 papers and answer keys, IIT Madras
- GATE 2027 Question Paper Pattern, IIT Madras
- GATE 2027 Data Science & Artificial Intelligence (DA) syllabus (PDF), IIT Madras
- Success Tracker GATE DA PYQ archive (question tags used for The ML Hub's subject, topic and question-type counts)