The rise of Data Science and Artificial Intelligence has been powered through areas like Machine Learning. As per a Forbes survey, 50% of companies plan to increase their spending on AI and Machine Learning in 2021. From Tech giants such as Google and Microsoft to leading startups like Uber and Slack, all are leveraging Machine Learning to understand data, solve real-world problems and drive future growth.
With such rapid adoption, Machine Learning is emerging as the most valuable Data Science skill. Consequently, the demand for Data Scientists with expertise in Machine Learning is at an all-time high. So, if you are applying for a Data Scientist position, you must prepare some common Machine Learning questions for the interview.
With a plethora of information available, preparing ML interview questions can be quite overwhelming. To streamline this for you, we have prepared a list of 20 Machine Learning questions (with answers) you are highly likely to be asked in a Data Scientist interview. This article covers basic, intermediate, and advanced-level machine learning interview questions. Practice these questions well and you’ll be all set for your Data Scientist interview.
Before we move on to the list of machine learning interview questions, if you are someone planning to transition to Data Science, check out our PGP in Data Science & AI course. With this program, you can successfully transition to a career in Data Science and AI in just 15 months.
Machine Learning Interview Questions
1. Name the different types of machine learning algorithms?
The three most important types of ML algorithms are:
- Supervised learning
- Unsupervised learning
- Reinforcement learning
2. What do you understand by data standardization?
Standardization in machine learning means that the features are rescaled such that they have a mean of 0 and a standard deviation of 1. Data Standardization helps compare the features with different units.
3. What is R2?
R2 is the coefficient of determination. It is a numerical value that measures the goodness of fit. With the value of R2, we can determine how close the actual and predicted outputs are in a regression problem. R2 = 1 indicates the perfect fit.
4. Explain conditional probability.
As per conditional probability – An event E will occur if some other event F has already occurred, that is, P(E|F) = P(EF) / P(F) where,
P(EF) = Probability that both events will occur
P(F) = Probability that event F will occur.
5. Why is the kernel trick used?
The kernel trick is used when we want to separate data. It is used to map data to a higher-dimensional space. It helps to avoid computing new coordinates of data points in this space.
6. What is clustering?
Clustering or cluster analysis is defined as the process of grouping observations or data points into two or more groups known as clusters. These clusters are created on the basis of similar features of data points.
Some popular clustering methods include k-means clustering, mean-shift clustering, agglomerative clustering, spectral clustering, affinity propagation, and DBSCAN.
7. Name some performance measures for regression problem
- MAE
- MSE
- RMSE
- R Squared (R²) and Adjusted R Squared
8. When would you use GD over SDG?
GD minimizes the error function better than SGD. However, SGD converges faster in a large dataset. Thus, GD is used for small datasets while SGD for larger ones.
9. Name the ways to control overfitting?
- Cross-validation
- Train with more data
- Remove features
- Early stopping
- Regularization
- Ensembling
10. Explain a ROC curve?
A Receiver Operator Characteristic (ROC) curve is used to depict the diagnostic ability of binary classifiers. It is widely used across areas such as medicine, radiology, natural hazards, and machine learning.
11. How are Random Forest Models different from ExtraTrees?
In the case of random forests, the locally optimal feature/split combination for each feature under consideration is computed, but in the case of extra trees, a random value is selected for the split.
12. What are other distance measures apart from simple euclidean distances?
- Manhattan distance
- Minkowski distance
13. What impurity measures are used to build decision-tree and related models in python’s scikit-learn library?
Common measures of impurity are Gini, Entropy, and Misclassification.
14. What are the disadvantages of decision trees?
The major disadvantage of decision trees is that they are prone to overfit. However, this can be addressed by ensemble methods such as random forests or boosted trees.
15. What is the difference between KNN and K-Means?
In K-Nearest Neighbours or KNN, classifies the labelled data points based on the distance of the point from the nearest points. We have to provide these labelled data to the model.
In case of K-Means clustering, we have to provide the model with unlabelled data. K-Means then classifies points into clusters based on the mean of the distance between different points.
16. How are regression trees built?
A regression tree is built through the process of binary recursive partitioning. The data is split into partitions which are then split into smaller groups, repeatedly as the method moves up each branch.
17. What is Cosine Similarity?
Cosine similarity is used to measure the similarity between two documents irrespective of their size. It measures the cosine of the angle between two non-zero vectors of an inner product space. The cosine of 0° is 1. It is less than 1 for any angle in the interval (0,π] radians.
18. What is regularization? Give some examples of regularization techniques?
Regularization is a technique that helps improve the validation score. Sometimes even at the cost of reducing the training score. Some popular regularization techniques are:
- L1 Regularization
- L2 Regularization
- Dropout
- Early stopping
19. What is an imbalanced dataset? Can you list some ways to deal with it?
An imbalanced dataset is one with different proportions of target categories. Some ways to deal with imbalanced datasets include:
- Oversampling or Undersampling
- Data Augmentation
- Using appropriate metrics
20. What are the advantages of neural networks?
The two key advantages of neural networks are:
- They lead performance breakthroughs in the case of unstructured datasets such as images, audio, and video.
- They are incredibly flexible. It allows them to learn any kind of patterns.
We hope these Machine Learning questions help you crack Data Science interviews. If you want us to cover interview questions any specific Data Science topics, let us know in the comments below.
1 comment