Artificial Intelligence & Data Science Interview Questions & Answers

Freshers in AI and Data Science are interviewed for roles such as data analyst, junior data scientist, machine learning engineer trainee and AI developer. The questions start with simple definitions (what is a data point, what is missing data) and quickly move to judgement: why a model is overfitting, how to choose a metric and how to clean messy data. This page lists the 50 common questions with plain answers and then adds what interviewers really look for: the round structure, a revision list, two worked examples (model metrics and overfitting) and the mistakes that cost candidates offers. Use it alongside your own mini-projects, because every interviewer will ask you to walk through one.

How a AI & Data Science fresher interview usually runs

Online test — aptitude, SQL or Python questions and sometimes a small data-cleaning task.
Technical interview — statistics, machine learning basics, Python libraries and a deep-dive into your project.
Case or problem round — a business problem such as predicting churn, where you describe data, features, model and metric.
HR round — communication, teamwork and how you keep learning.

Revise these 8 topics first

Statistics: mean, median, mode, variance, standard deviation, normal distribution and correlation vs causation.
Data cleaning: missing values, outliers, duplicates and encoding categorical data.
Core ML ideas: regression vs classification, supervised vs unsupervised, and train-validation-test split.
Evaluation metrics: accuracy, precision, recall, F1 score and when each is misleading.
Overfitting and underfitting: bias-variance, regularisation and cross-validation.
Algorithms: linear and logistic regression, decision tree, random forest, K-Means and K-NN.
Python tools: NumPy, pandas, matplotlib and scikit-learn.
SQL: joins, GROUP BY and a window function.

Showing the first 50 questions free. The complete 250-question bank is available as a PDF below.

Artificial Intelligence & Data Science — Q1 to Q50

Q1What is Artificial Intelligence?

AnswerArtificial Intelligence, or AI, is a field of computer science focused on developing systems that can perform tasks that normally require human intelligence, such as learning, reasoning, pattern recognition, decision-making, language understanding, and perception.

Q2What is Data Science?

AnswerData Science is an interdisciplinary field that uses statistics, mathematics, programming, machine learning, and domain knowledge to extract useful insights and make decisions from data.

Q3What is the difference between AI, Machine Learning, and Data Science?

AnswerAI is the broader field of creating intelligent systems. Machine Learning is a subset of AI where systems learn patterns from data instead of being explicitly programmed for every situation. Data Science focuses on collecting, processing, analyzing, and interpreting data to generate insights and support decision-making.

Q4What is Machine Learning?

AnswerMachine Learning is a method of developing computer systems that learn patterns from data and use those patterns to make predictions or decisions. The model improves its performance through suitable training data and learning algorithms.

Q5What are the main types of Machine Learning?

AnswerThe main types are supervised learning, unsupervised learning, and reinforcement learning. Supervised learning uses labeled data, unsupervised learning identifies patterns in unlabeled data, and reinforcement learning learns through interactions with an environment using rewards or penalties.

Q6What is supervised learning?

AnswerSupervised learning trains a model using input data along with known target outputs. The model learns the relationship between inputs and outputs and then uses that relationship to make predictions on new data.

Q7What is unsupervised learning?

AnswerUnsupervised learning works with data that does not have predefined target labels. The algorithm attempts to discover underlying patterns, groups, structures, or relationships within the data.

Q8What is reinforcement learning?

AnswerReinforcement learning is a learning approach in which an agent interacts with an environment and receives rewards or penalties based on its actions. The objective is to learn a strategy that maximizes cumulative reward over time.

Q9What is a dataset?

AnswerA dataset is a structured collection of data used for analysis, reporting, or machine-learning tasks. It may contain rows representing observations and columns representing features or variables.

Q10What is a feature in Machine Learning?

AnswerA feature is an input variable used by a machine-learning model to make a prediction or decision. For example, in a house-price model, area, number of rooms, location, and age of the property can be features.

Q11What is a target variable?

AnswerThe target variable is the output that a supervised machine-learning model is designed to predict. For example, if we are predicting house prices, the house price is the target variable.

Q12What is a label?

AnswerA label is the known output associated with a training example in supervised learning. For example, in an email-classification dataset, "spam" or "not spam" can be labels.

Q13What is a data point?

AnswerA data point is an individual observation or record in a dataset. For example, one customer's transaction record can represent one data point.

Q14What is structured data?

AnswerStructured data is organized in a predefined format, commonly rows and columns in databases or spreadsheets. Examples include customer records, sales transactions, and employee databases.

Q15What is unstructured data?

AnswerUnstructured data does not follow a fixed tabular structure. Examples include images, videos, audio recordings, emails, documents, and free-form text.

Q16What is semi-structured data?

AnswerSemi-structured data does not follow a rigid table structure but contains organizational elements such as tags or keys. JSON and XML are common examples.

Q17What is data preprocessing?

AnswerData preprocessing is the process of preparing raw data for analysis or machine learning. It can include handling missing values, removing duplicates, correcting inconsistencies, encoding categorical variables, scaling numerical features, and handling outliers.

Q18Why is data preprocessing important?

AnswerReal-world data is often incomplete, inconsistent, noisy, or differently formatted. Proper preprocessing improves data quality and can significantly affect the accuracy, stability, and reliability of machine-learning models.

Q19What is missing data?

AnswerMissing data occurs when a value is unavailable for one or more variables in a dataset. Depending on the situation, it may be handled by removing records, filling values using appropriate statistical methods, or using models capable of handling missing values.

Q20What is data cleaning?

AnswerData cleaning involves identifying and correcting errors, inconsistencies, duplicates, missing values, incorrect formats, and other quality issues in a dataset before analysis or modeling.

Q21What is an outlier?

AnswerAn outlier is an observation that differs significantly from the general pattern of the data. Outliers may represent genuine unusual cases, measurement errors, data-entry errors, or other special conditions, so they should be investigated rather than automatically removed.

Q22What is exploratory data analysis?

AnswerExploratory Data Analysis, or EDA, is the process of examining and summarizing a dataset to understand its structure, distributions, relationships, trends, and potential problems. It commonly uses statistical measures and visualizations.

Q23What is correlation?

AnswerCorrelation measures the degree to which two variables are associated with each other. A positive correlation means they tend to increase together, while a negative correlation means one tends to increase as the other decreases. Correlation alone does not prove causation.

Q24What is the difference between correlation and causation?

AnswerCorrelation means two variables show an association, whereas causation means a change in one variable directly contributes to a change in another under an appropriate causal interpretation. Two variables can be correlated because of another underlying factor without one causing the other.

Q25What is mean, median, and mode?

AnswerMean is the arithmetic average of values. Median is the middle value when observations are arranged in order, and mode is the value that occurs most frequently. The median can be more representative than the mean when the data contains significant outliers.

Q26What is standard deviation?

AnswerStandard deviation measures how much the values in a dataset typically vary around the mean. A higher standard deviation indicates greater dispersion, while a lower value indicates that observations are more closely concentrated around the mean.

Q27What is variance?

AnswerVariance measures the average squared deviation of observations from their mean. Standard deviation is the square root of variance and is expressed in the same units as the original variable.

Q28What is probability in Data Science?

AnswerProbability is a mathematical measure of the likelihood of an event occurring. It is fundamental to statistics, predictive modeling, uncertainty estimation, Bayesian methods, and many machine-learning algorithms.

Q29What is regression?

AnswerRegression is a supervised learning approach used to predict a numerical or continuous target variable. For example, it can be used to predict house prices, sales revenue, temperature, or vehicle fuel consumption.

Q30What is classification?

AnswerClassification is a supervised learning task in which the model predicts a category or class. Examples include classifying emails as spam or not spam and predicting whether a transaction is potentially fraudulent.

Q31What is the difference between regression and classification?

AnswerRegression predicts a continuous numerical output, while classification predicts a discrete class or category. For example, predicting salary is a regression problem, whereas predicting whether an employee will leave a company is a classification problem.

Q32What is linear regression?

AnswerLinear regression models the relationship between input variables and a continuous target using a linear function. The model estimates coefficients that minimize an appropriate measure of prediction error on the training data.

Q33What is logistic regression?

AnswerLogistic regression is a supervised learning algorithm commonly used for classification. It estimates the probability of a class using a logistic function and can then assign a class based on a selected decision threshold.

Q34What is a decision tree?

AnswerA decision tree is a supervised learning model that makes predictions through a sequence of decision rules. Internal nodes represent conditions on features, branches represent outcomes, and leaf nodes represent predictions.

Q35What is Random Forest?

AnswerRandom Forest is an ensemble learning method that combines predictions from multiple decision trees. The trees are trained using variations of the data and features, and their combined prediction generally provides better robustness than relying on a single tree.

Q36What is clustering?

AnswerClustering is an unsupervised learning technique that groups similar data points together. For example, customer records can be clustered according to purchasing behavior without predefined customer categories.

Q37What is K-Means clustering?

AnswerK-Means is a clustering algorithm that divides data into a specified number of clusters. It assigns observations to clusters based on distance from cluster centroids and iteratively updates those centroids until the clustering stabilizes according to the algorithm's stopping criteria.

Q38What is overfitting?

AnswerOverfitting occurs when a model learns the training data too closely, including noise or accidental patterns, and therefore performs poorly on unseen data. A model can have very good training performance but poor generalization.

Q39What is underfitting?

AnswerUnderfitting occurs when a model is too simple or insufficiently trained to capture important patterns in the data. As a result, it performs poorly on both training and unseen data.

Q40How can overfitting be reduced?

AnswerOverfitting can be reduced through techniques such as using more representative data, regularization, cross-validation, feature selection, reducing model complexity, pruning decision trees, dropout in neural networks, and early stopping where appropriate.

Q41What is train-test splitting?

AnswerTrain-test splitting divides the available dataset into separate subsets. The training set is used to develop the model, while the test set is kept separate to evaluate how well the final model generalizes to unseen data.

Q42What is a validation dataset?

AnswerA validation dataset is used during model development to compare models, tune hyperparameters, or make design decisions without using the final test set. A common setup is training, validation, and test datasets.

Q43What is cross-validation?

AnswerCross-validation is a model-evaluation technique in which the data is divided into multiple parts and the model is trained and evaluated across different combinations of those parts. It provides a more robust estimate of model performance than relying on a single split.

Q44What is accuracy in Machine Learning?

AnswerAccuracy is the proportion of predictions that are correct among all evaluated predictions. It can be useful when classes are reasonably balanced, but it may be misleading for highly imbalanced datasets.

Q45What is a confusion matrix?

AnswerA confusion matrix summarizes classification results by showing true positives, true negatives, false positives, and false negatives. It helps understand not only how many predictions were correct but also the types of errors made by the model.

Q46What are precision and recall?

AnswerPrecision measures the proportion of predicted positive cases that are actually positive. Recall measures the proportion of actual positive cases that the model successfully identifies. The appropriate metric depends on the business or application requirement.

Q47What is F1-score?

AnswerF1-score is the harmonic mean of precision and recall. It provides a single metric that balances both measures and can be useful when false positives and false negatives are both important.

Q48What is Python's role in AI and Data Science?

AnswerPython is widely used because it has a simple syntax and a large ecosystem of libraries for data manipulation, visualization, statistics, machine learning, and deep learning. Common libraries include NumPy, pandas, Matplotlib, scikit-learn, and deep-learning frameworks such as TensorFlow and PyTorch.

Q49What are NumPy and pandas used for?

AnswerNumPy provides efficient numerical arrays and mathematical operations and is widely used as a foundation for scientific computing in Python. pandas provides data structures such as Series and DataFrame and is commonly used for data cleaning, transformation, analysis, and manipulation.

Q50A machine-learning model has very high training accuracy but poor test accuracy. What would you investigate?

AnswerI would first suspect overfitting and check the training and test data distributions, feature quality, model complexity, and data leakage. I would evaluate the model using appropriate cross-validation, review preprocessing and feature selection, and consider regularization, simpler models, more representative data, or other techniques to improve generalization.

Worked examples interviewers like

Example 1: “A model predicts a disease. Out of 200 patients: TP = 40, FP = 10, FN = 20, TN = 130. Calculate the metrics.”

Accuracy = (TP + TN) / total = (40 + 130) / 200 = 0.85.
Precision = TP / (TP + FP) = 40 / 50 = 0.80.
Recall = TP / (TP + FN) = 40 / 60 = 0.667.
F1 = 2 × 0.80 × 0.667 / (0.80 + 0.667) = about 0.727.

Add the insight: for a disease test, missing a sick patient (low recall) is usually costlier than a false alarm, so recall deserves attention even when accuracy looks good.

Example 2: “Your model scores 98% on training data but 78% on test data. What is happening and what will you do?”

This is overfitting: the model has memorised the training data instead of learning general patterns.

Check for data leakage (test information accidentally used in training).
Use cross-validation to get a more reliable estimate.
Simplify the model, limit tree depth or add regularisation.
Get more data or use augmentation, and remove noisy features.

Common mistakes that cost candidates the job

Quoting accuracy alone on an imbalanced data set. Always ask about class balance.
Cannot explain the algorithm used in their own project and why it was chosen.
Calling every problem “AI” and ignoring the simple baseline model.
Ignoring data cleaning, which takes most of the real working time.
Mixing correlation with causation in a business answer.

Frequently asked questions

Do I need deep learning for a fresher interview?

Not always. Most entry roles focus on statistics, Python, SQL and classical ML. Know the basic idea of neural networks and one project if you can.

Which language should I use?

Python is the standard. Add SQL, because nearly every data role uses it.

How many projects do I need?

Two well-understood projects are better than five copied ones. Be ready to explain data, method, results and limits.

Keep preparing

→ B.Tech Computer Science (CSE) Interview Q&A → B.Tech Information Technology (IT) Interview Q&A → B.Tech ECE Interview Q&A → B.Tech Biomedical Engineering Interview Q&A → All branch-wise interview Q&A (hub)

📚 Buy Full Artificial Intelligence & Data Science Interview Q&A PDF — ₹49

You’ve read the first 50 questions here. Get the complete 250-question bank PDF with all questions and clear answers. One-time price: ₹49.

🛒 Buy PDF – ₹49