Data Science Interview Questions for Freshers: Python, Statistics & ML

A list is mutable — you can add, remove, or change elements after creating it. A tuple is immutable — once created, it can't be changed. Lists use square brackets, tuples use parentheses.

Interview tip: mention that tuples are generally faster and used when data shouldn't change, such as coordinates or fixed configuration values.

A DataFrame is a two-dimensional, table-like structure in Pandas, with labeled rows and columns — similar to a spreadsheet. It's the core structure used to load, clean, and analyze datasets in Python.

First check how much data is missing and why, since that shapes the right approach. Common options: drop rows with missing values, fill them with the mean or median, or use a business-appropriate default.

Interview tip: avoid saying "I'd just drop them" without qualification — explain that you'd first check what percentage is missing and whether the column is important.

NumPy arrays are faster and more memory-efficient for numerical operations, and they support element-wise operations directly — something Python lists don't do without a loop.

The groupby() function splits data into groups based on a column, then applies an aggregate function like sum, mean, or count to each group.

A function is a reusable block of code that performs a specific task. In a data project, functions help avoid repeating the same cleaning or analysis steps across multiple datasets, and make your code easier to test and explain.

Mean is the average of all values. Median is the middle value once sorted. Mode is the most frequently occurring value. For data like 10, 20, 20, 30, 500 — the mean gets pulled up sharply by the outlier, while the median stays a more honest representation of the typical value.

Standard deviation measures how spread out values are from the mean. A low standard deviation means values cluster close to the average; a high one means they're spread widely. It matters because two datasets can have the same average but very different levels of consistency.

Correlation measures how strongly two variables move together — positive, negative, or none. It does not imply causation. Ice cream sales and drowning incidents may correlate because both rise in summer, but one doesn't cause the other.

Interview tip: always mention the "correlation isn't causation" distinction — interviewers listen for it specifically.

An outlier is a value unusually different from the rest of the dataset. Before removing one, investigate why it exists — it could be a data-entry error, or it could be a genuine, important event. Don't treat "unusual" as automatically "wrong."

Probability measures how likely an event is, expressed between 0 and 1. In practice, it underlies things like the confidence behind a model's prediction, or the likelihood of a customer converting based on historical patterns.

A symmetric, bell-shaped distribution where most values cluster around the mean, with fewer values further away in either direction. Many real-world measurements — heights, test scores, measurement errors — approximate this shape, which is why it's foundational to a lot of statistical reasoning.

Supervised learning trains a model on labeled data, where the correct answer is already known — like predicting house prices from past sales. Unsupervised learning works with unlabeled data to find patterns or groupings on its own, like customer segmentation.

Overfitting happens when a model learns the training data too precisely, including its noise, and then performs poorly on new, unseen data. It's like memorizing exam answers instead of understanding the subject.

Ways to reduce it: use more training data, simplify the model, or use techniques like cross-validation to check performance on data the model hasn't seen.

Classification predicts a category — will this customer churn, yes or no. Regression predicts a continuous number — what will next month's revenue be. The type of question you're answering determines which one applies.

By evaluating it on data it wasn't trained on, using metrics appropriate to the problem — accuracy for balanced classification, precision and recall when one type of error matters more than another, or RMSE for regression. High accuracy on training data alone tells you very little.

K-Means groups data points into a chosen number of clusters based on similarity, by repeatedly assigning points to the nearest cluster center and adjusting those centers until the groupings stabilize. It's commonly used for customer segmentation.

To honestly evaluate how a model performs on data it hasn't seen before. Testing on the same data you trained with gives a falsely optimistic picture, since the model has effectively already "seen the answers."