Cracking a data science interview requires a thorough knowledge of the field as well as domain expertise. As data science is one of the fastest growing fields today, many fresh graduates and young working professionals aspire to build a career in this field. This is the reason why offline and online data science courses are gaining more popularity these days. The right course can prepare you for your dream job, but you should also make some additional effort to excel in a data science interview.

You must be aware of what type of questions can be asked in an interview and which soft skills matter the most. You can find the list of the Top 50+ data science interview questions and their answers further in this article:

**1. What are the different steps involved in data cleaning?**

**Ans.** Data Cleaning is an important part of the data analysis process as it helps improve the accuracy of data, making it more reliable for the decision-making process. The following are some important steps involved in data cleaning:

- Checking the data quality
- Removing irrelevant and duplicate data points
- Looking for structural errors and correcting them
- Removing or changing the outliers
- Dealing with the missing data points

**2. Why is Data Cleaning Important?**

**Ans.** Data Cleansing refers to the process of removing or changing incorrect or irrelevant values from data. To improve the accuracy of data and the productivity of business processes, it is important to check for its quality before performing data analysis. The large real-world data sets, when captured, are often inconsistent and can have errors or missing values. To filter useful data from the raw data, the data cleansing process becomes essential.

**3. Explain under-fitting and overfitting.**

**Ans.** Underfitting refers to a situation where the model is very simple and cannot find the correct relationship between the data. As a result, it doesn’t perform well on the test data. This generally happens due to low variance and high bias in the model. Underfitting is more likely to occur in the case of linear regression.

Overfitting is the situation where the model is only suitable for the sample data and doesn’t perform well in general. It is unable to provide any result if any new data is provided to this model. The main reason behind overfitting is high variance and low bias. Overfitting is more likely to occur in the case of decision trees.

**4. Define linear and logistic regression.**

**Ans.** Linear Regression is a simple technique that you can use to calculate the value of dependent variable Y on the basis of the value of independent variable X. Here, Y is called the criterion variable and X is the predictor variable. The technique used the simple equation Y = mX + C to find the line of best fit, where m is the slope and c is the standard error.

Logistic Regression is used when you need to predict a variable that is binary in nature. In this case, the number of possible outcomes is limited. You can either get 0 or 1, where 1 generally denotes the occurrence and 0 denotes the non-occurrence of an event.

**5. What are the types of biases that occur during sampling?**

**Ans.** Generally, the following three types of biases can occur during sampling:

**Self-Selection Bias:**This occurs when the participants select themselves for an analysis.**Under Coverage Bias:**This occurs when very few samples are selected from a large population.**Survivorship Bias:**This occurs when analysts focus on elements that passed a selection process while ignoring the ones that did not.

**6. Define Eigenvalues and Eigenvectors.**

**Ans.** Eigenvectors are the unit or column vectors whose magnitude is equal to 1. These are also known as the right vectors and use to define the directions in which a linear transformation is moving. Eigenvalues are the coefficients that are multiplied with the eigenvectors to give these vectors different values for the magnitude or distance.

**7. What are the steps involved in making a Decision Tree?**

**Ans.** You need to follow the below-mentioned steps while making a decision tree:

- Determining the root of the tree step
- Calculating entropy for the classes step
- Calculating entropy after splitting for each attribute
- Calculating the gain of information for each split
- Performing the split and then further split steps
- Completing the decision tree model

**8. Which Python Libraries can be used to plot data?**

**Ans.** Data Scientists commonly use these libraries to plot data in Python:

**Seaborn:**For pair plots to determine the relationship between different variables in a data set.**Matplotlib:**For different types of plots, such as histograms, bar graphs, pie charts, area plots, scatterplots, etc.**Ggplot:**For bar and line plots, density plots, error bars, histograms, violin plots, box plots, etc.

**9. What is Reinforcement Learning?**

**Ans.** Reinforcement Learning is a type of machine learning that works on a feedback-based mechanism. In this, a model is rewarded when it makes desired observations and punished when it makes the undesired ones. The model tries to achieve the maximum rewards by making more desired actions and learning from its mistakes when punished.

**10. What do you mean by cross-validation?**

**Ans.** The cross-validation technique is used when the major goal is to ensure the reliable performance of a machine-learning model. In this method, the model is trained and tested on different samples of data in order to make sure that the model performs well on unknown datasets. The entire training data is divided into different groups and then the model is tested against all these groups in rotation.

**11. What steps do you follow to build a random forest model?**

**Ans.** A random forest can be defined as a group of multiple decision trees. The steps included in building a random forest are mentioned below:

- Out of the total ‘m’ features, you have to select ‘k features randomly. Here, k<
- Use the best split point along ‘k’ features to calculate node D.
- Split this node further into daughter nodes using the best split.
- Keep repeating these steps until you finalize the leaf nodes.
- To build the random forest, repeat the above 4 steps ‘n’ times, where ‘n’ is the number of decision trees you want to create.

**12. Define Uniform and Skewed Distribution.**

**Ans.** In a uniform distribution, the probability of all the outcomes is equally likely. On the other hand, skewed distribution refers to the probability distribution where most data points lie to the right or left of the center, i.e. the probability of all outcomes is not equally likely.

**13. What are recommender engines?**

**Ans.** Recommendation engines are systems built using data science and Ml techniques that recommend relevant products or services to customers on the basis of their interests and past purchasing behaviors. Businesses use these systems to provide their customers with more personalized experiences.

**14. Define a Gradient and a Gradient Descent.**

**Ans.** A gradient can be defined as a measure of how much the output changes with a slight change in the input. In technical words, it is the measure of change in weights w.r.t. the change in errors.

Gradient Descent is an algorithm that deals with the minimization of the activation function. It is used to find the line of best fit with the minimum possible number of iterations.

**15. What is the difference between correlation and covariance?**

**Ans.** Correlation measures the relationship between two variables. If two variables are directly proportional (they increase or decrease together), there is a positive correlation. Similarly, there is a negative correlation when the variables are indirectly proportional (one increases when the other decreases).

Covariance is the measure of how much two variables change together in a cycle. A high covariance explains that the two variables are strongly related whereas a low covariance refers to a weaker relationship.

**16. What is the p-value?**

**Ans.** P-value is used to determine the strength of your outcomes when performing a hypothesis test. The claim which is to be tested is known as the null hypothesis. A p-value less than 0.05 indicates that the null hypothesis is false and should be rejected. On the other hand, a p-value of more than 0.05 indicates that the null hypothesis is true and can be accepted. An exact p-value of 0.05 indicates that the null hypothesis can either be true or false.

**17. What is a normal distribution?**

**Ans.** A normal distribution, also known as a Gaussian distribution, is a type of probability distribution in which most of the values are close to the mean. The characteristics of such a distribution are listed below:

- The mean, mode, and median of the distribution are the same.
- It has a bell-shaped curve.
- The total area covered under the curve is equal to 1.
- The number of values to the right and the left of the center are equal.

**18. Define bias-variance tradeoff.**

**Ans.** When the algorithm used in your model is over-simplified, the error is known as a bias and variance is the error that occurs due to the over-complexity of your ML algorithm. The bias-variance tradeoff refers to maintaining a balance between the bias and variance in an ML model. The bias decreases with the increase in variance and vice versa. The total error is calculated as a sum of bias, the square of variance, and the irreducible error.

**19. What do you understand by exploding gradients?**

**Ans.** Exploding gradients can be defined as a problematic scenario where a large number of error gradients get multiplied and become huge over time. In such a case, the weights’ value can overflow and result in NaN values. As a result, the model is unable to learn from the training data as it becomes unstable.

**Also Read:** Crack Your Data Science Interview with These Top 15 Questions for 2023

**20. When is re-sampling done?**

**Ans.** Re-sampling refers to the creation of new samples on the basis of one pre-observed sample. Re-sampling is done for the following purposes:

- To validate models using randomly selected subsets.
- To estimate sample statistics accuracy with the subsets of accessible datasets.
- To substitute the data point labels while you perform significance tests.

**21. What is the Confusion Matrix and How is it Used to Calculate Accuracy?**

**Ans.** A confusion matrix is a 2*2 matrix (2 rows and 2 columns) having 4 outputs provided by a binary classifier. Using a confusion matrix, one can derive several measures including accuracy, error rate, sensitivity, precision, specificity, and recall. To calculate the accuracy of a model, the following formula is used:

- (TN+TP)/(TN+TP+FN+FP), where TN is True Negative, TP is True Positive, FN is False negative, and FP is False Positive.

**22. How to avoid the over-fitting of your model?**

**Ans.** Over-fitting refers to a condition where there are a large number of variables for small amounts of data. This type of model performs well only on the sample training data and fails for other data sets. To avoid over-fitting, techniques like cross-validation (K folds) and regularisation (Lasso Regression) can be used.

**23. What do you understand by support vectors in SVM?**

**Ans.** In SVM, support vectors are the data points that lie close to the hyperplane. These are among the important factors that influence the orientation and position of the hyperplane. You can maximize the margin of the classifier under these support vectors. If these vectors are deleted, the position of the hyperplane is changed.

**24. What is a computational Graph?**

**Ans.** A computational graph, also called a dataflow graph, in Tensorflow consists of a network of nodes where each node is operational. In this graph, the edges represent tensors and the nodes represent operations. Everything in TensorFlow (the famous deep learning library) works on the basis of a computational graph.

**25. Explain the fundamentals of Neural networks.**

**Ans.** The Neural Network in deep learning is a network of artificial neurons that are designed to mimic the human brain. It learns from the patterns in data provided to it and uses this knowledge to predict outcomes for new data without any human interference. Neural networks are made up of different layers as listed below:

- An input layer to receive the input.
- One or multiple hidden layers to detect patterns and combine outputs from previous layers.
- An output layer, which is used to make the final prediction.

**26. Define Survivorship Bias.**

**Ans.** Survivorship bias is a type of error that occurs during sampling. It can be defined as a logical error that occurs when you focus on aspects that passed some process while overlooking or ignoring the ones that did not. This bias can result in wrong predictions.

**27. What are confounding variables?**

**Ans.** Confounding variables or confounders are a kind of extraneous variables that affect other dependent or independent variables in such a way that produces a distorted association between these variables. These variables confound or confuse the true relationship between two variables that are associated but not causally related to each other.

**28. Explain MSE and RMSE in a linear regression model.**

**Ans.** MSE (Mean Squared Error) is used to determine how close the best-fit line is to the actual data. It is calculated as the square of the difference between the actual value and the predicted value. RMSE (Root Mean Squared Error) is used to test the performance of a linear regression model and to study the data spread around the best-fit line. The formula to calculate RMSE is the square root of MSE.

**29. How can you use data visualizations effectively in data science projects?**

**Ans.** Data visualization is very helpful in creating interactive reports and dashboards that make complex data easier to understand for the human mind. Data visualizations can be used to explore and analyze data effectively by understanding the patterns and trends in past data. Some most commonly used visualization tools for data science projects include MS Excel, Tableau, QlikView, etc.

**30. Define root cause analysis.**

**Ans.** Root Cause Analysis can be defined as the process of tracking the occurrence of an event and the causes that lead to it. You generally do it in the case of software malfunctions. Data scientists can use root cause analysis to help a business understand the factors and semantics behind certain outcomes.

**31. What are the drawbacks of a linear model?**

**Ans.** Some most considerable drawbacks of a linear model are as follows:

- The assumptions that this model makes regarding the linearity of the errors.
- This model is not useful for count outcomes or binary outcomes.
- A linear model is unable to solve some overfitting problems.
- The model assumes that there is no multicollinearity in data.

**32. Why is A/B testing important?**

**Ans.** A/B testing aims to pick the best variant among the two hypotheses. This type of testing is used for the responsiveness of web pages or applications, banner testing, redesigning the landing pages, performance of marketing campaigns, etc. In the first step, you have to confirm the conversion goal and then use statistical analysis to understand which alternative gives better performance for the conversion goal.

**33. What is the law of large numbers?**

**Ans.** According to the ‘law of large numbers,’ if you repeat an experiment independently a large number of times, the average individual result comes out to be close to the expected value. Moreover, the law says that the standard deviation and sample variance also converge toward the expected value.

**34. What are the feature selection methods for selecting the right variables?**

**Ans.** The feature selection methods can be divided into two types as follows:

- Filter Methods, which include ANOVA, linear discrimination analysis, and chi-square method.
- Wrapper Methods, which include forward selection, backward selection, and recursive feature elimination.

**35. How do you treat outlier values?**

**Ans.** Outlier values can be treated by replacing the values with mode, mean, or a cap-off value. Another way is to remove all the rows containing outliers if they make up only a small proportion of the entire dataset. You can also perform data transformation on the outlier values.

**36. Differentiate between ‘long’ and ‘wide’ format data.**

**Ans.** In long-format data, you have as many rows as the number of attributes for each data point. Here, every row contains a particular attribute’s value for a given data point. On the other hand, wide format is a type of data format where you have a single row containing every data point with multiple columns holding the values of multiple attributes.

**Also Read:** The Top 20+ Data Science Skills You Need by 2023

**37. Is a random forest better than multiple decision trees?**

**Ans.** Yes, a random forest is better than multiple decision trees. It is because random forests are much more accurate, robust, and less prone to overfitting. The model contains multiple weak decision trees and ensures that all of these trees learn strongly.

**38. Explain the difference between append and extend methods.**

**Ans.** The append () method in Python is used to add items to an already existing list. Extend () method is used to add multiple individual elements at the end of the original list.

**39. When you exit Python, is all the memory de-allocated?**

**Ans.** No, all the memory doesn’t necessarily get de-allocated when you exit Python. It is because the objects that have circular references are not always free when Python exits.

**40. How is time series different from other regression methods?**

**Ans.** Time series is extrapolation whereas the other regression problems are interpolation. Time series can be defined as an organized chain of data and it predicts the next values in the sequence. Time series can be used along with other series that can occur simultaneously. Talking about the regression methods, these can be applied with non-ordered sequences (called features) as well as time series problems. When you make a projection, new feature values are presented and regression calculates the values for the target variable.

**41. Why is dimensionality reduction important?**

**Ans.** Dimensionality Reduction, as the name suggests, refers to the process of reducing the dimensions of a data set in order to convey the same information more concisely. This method is very effective in saving storage space by compressing the data. It also reduces the computation time and helps in removing redundant features.

**42. Can you get a better predictive model by treating a categorical variable as a continuous variable?**

**Ans.** Not necessarily, but yes, treating a categorical variable as a continuous variable might be helpful in some cases. However, it is not a preferred approach as it only works when the variables in question have an ordinal nature.

**43. Explain box-cox transformation.**

**Ans.** A Box-Cox transformation is a method to normalize variables as it is essential for various statistical techniques. If your data is not normal, you can apply the box-cox transformation to be able to run more tests. This transformation can help you improve the accuracy of your linear regression models.

**44. Define a hyperbolic tree.**

**Ans.** A hyperbolic tree, also known as a hypertree, is a type of graph drawing and information visualization method that takes inspiration from hyperbolic geometry.

**45. What are some drawbacks of data visualization?**

**Ans.** Though data visualization is very effective in making data easy to understand for the human mind, it has some downsides as well. Some of its most important advantages are as follows:

- The estimations of visualizations are not always accurate.
- Different people may interpret data visuals differently.
- Improperly designed visualizations may cause confusion.

**46. Mention data mining packages in R.**

**Ans.** Some popular data mining packages in R include:

- Ggplot2 for data visualization
- Dplyr for data manipulation
- Purrr for data wrangling
- Datapasta for data import
- Hmisc for data analysis

**47. How to maintain a deployed model?**

**Ans.** To improve the performance of a deployed model, it should be retrained after a while. After deployment, you should track the performance of the model and keep a record of its predictions compared with the true values. Later, the model should be retrained on new data. Moreover, you can perform root cause analysis to identify the cause for the wrong predictions.

**48. Explain collaborative filtering.**

**Ans.** Collaborative filtering is a technique through which you can filter out the points that a user may like based on the interests and reactions of other similar users. It first searches for a large group of people and then finds smaller subgroups of people having similar interests to a particular user.

**49. What do you mean by pruning in a decision tree?**

**Ans.** Pruning refers to the process of reducing the size of a decision tree by removing some nodes. Pruning is done because the decision trees made by the base algorithm are sometimes very large and complex, and can be prone to overfitting.

**50. Explain different assumptions used in linear regression? What happens if these assumptions are violated?**

**Ans.** The assumptions used in linear regression are listed below:

- The sample data represents the entire population.
- There is a linear relationship existing between the x-axis variable and the mean of the y-axis variables.
- For any X values, the residual variance remains the same.
- All the observations are independent of each other.
- For any value of X, Y is normally distributed.

**51. What is the difference between confidence intervals and point estimates?**

**Ans.** The confidence interval gives a range of values that are likely to contain the population parameter. It also tells you about the probability of a particular interval containing the population parameter.

A point estimate is the estimate of the population parameter provided by a certain value. Some popular techniques used to determine the population parameters’ point estimates include the Method of Moments and the Maximum Likelihood Estimator.

**52. What is Naive in Naive Bayes?**

**Ans.** The Naive Bayes works on the assumption that the absence or presence of a feature of a class is not related to the absence or presence of any other feature. It is called ‘naive’ because it uses certain assumptions that may or may not be true.

Edvancer is an online career-oriented learning platform that offers the following courses in data science:

- Certified Data Science Specialist
- IBM Professional Certificate in Data Science
- PG Program in Data Science – UPES
- Executive Program in Data Science for Managers

**1. What questions are asked in a data science interview?**

**Ans.** The types of questions asked in a data science interview depend on your experience level. If you have applied for a job as a fresher, you will be asked the most basic questions, such as important terms, definitions, differences, etc. However, if you are already experienced in the field, you will be asked advanced-level questions.

**2. What is the basic skill for data science?**

**Ans.** The basic skills required to become a data scientist include mathematics, programming, and statistics. Apart from these, you should have soft skills such as problem-solving, good communication, decision-making, etc.

**3. Do data scientists use Excel?**

**Ans.** Yes, Microsoft Excel is one of the most important and widely used tools by data scientists and analysts. It helps in preparing interactive reports, graphs, charts, etc.

Follow us on

Free Data Science & AI Starter Course