Machine Learning Interview Questions

Machine Learning interview demands a rigorous interview process where the interviewers judge the candidates on various aspects such as technical and programming skills, knowledge of methods, and clarity of basic concepts. This article from PythonGeeks is an attempt to assist you to crack the machine learning interviews at major product-based companies and start-ups and secure a higher-paying job through basic concepts.

Machine Learning Questions for Beginners

1. How do you perceive the term Machine learning?

Machine learning is the subset of Artificial Intelligence that tends to deal with system programming and even tries to automate data analysis to authorize computers to learn and act through experiences without being requiring programmers to explicitly program them.

For example, Engineers code Robots in such a way that they are able to perform the tasks on the basis of data they collect from sensors. They involuntarily learn programs from data and enhance themselves with experiences.

2. Differentiate between the two: inductive learning and deductive learning.

In inductive learning, the model tends to learn by examples from a set of observed instances in an attempt to draw a generalized conclusion. Whereas, on the other hand, in deductive learning, the model applies the conclusion at the beginning itself, and then draws the conclusion from it. To summarize it shortly:

• Inductive learning is the method that makes use of observations to draw conclusions.
• Deductive learning is the method that makes use of conclusions to form observations.

3. What is the fundamental difference between Data Mining and Machine Learning?

Data mining can be thought of as the process in which the structured data extensively attempts to abstract knowledge or fascinating unrecognized patterns. During this process, the data mainly uses machine learning algorithms.

Machine learning embodies the study, design, and development of effective algorithms which facilitate the ability of the processors to learn without requiring any explicit programming.

4. What is the meaning of Overfitting that you understand in Machine learning?

We can observe Overfitting in machine learning when a statistical model tends to describe random error or noise instead of the underlying relationship as the output. We usually observe Overfitting when a model is excessively complex indicating it has numerous features and data points. This situation occurs because of having too many parameters concerning the number of training data types in the given training dataset. The model where we can observe overfitting tends to exhibit poor performance.

5. When can you conclude that overfitting has occurred in the dataset?

The possibility of overfitting appears when the criteria that we use for training the model are not as per the criteria that we use to judge the efficiency of a model.

6. Suggest a method to avoid overfitting.

Overfitting occurs when we have a dataset that has smaller dimensions, and a model is trying to learn from this dataset. By making use of a large amount of data, we can avoid overfitting.

However, if we possess a smaller database and are bound to build a model based on that, then we can make use of a technique known as cross-validation. In this method, we provide a model with a dataset of known data on which the training data set runs and a dataset of unknown data against which we tend to test the model.

Cross-validation primarily focuses on defining a dataset to test the model in the training phase. If sufficient data is available, we can even make use of Isotonic Regression to prevent overfitting.

7. Differentiate supervised and unsupervised machine learning according to your understanding of both the learning methods.

In supervised machine learning, we tend to train the machine using labeled data. Then, we tend to provide a new dataset to the learning model so that the algorithm facilitates a positive outcome by analyzing the labeled data. For example, we first need to label the data which is a prerequisite to train the model while performing classification.

In unsupervised machine learning, we do not require labeled data or a supervisor to train the machine and let the algorithms make the decisions without any corresponding output variables of the previous dataset.

8. On the core level, what factors make Machine Learning different from Deep Learning?

Machine learning works predominantly with algorithms that programmers use to parse data, learn from that data, and then deploy their learnings to make informed decisions.

Deep learning is a subset of machine learning, which draws its inspiration from the structure of the human brain and is particularly useful in the arena of feature detection.

9. Differentiate KNN from K-means.

KNN or the full form, K nearest neighbors is a type of supervised learning algorithm which is beneficial for classification problems. In KNN, we provide a test sample as the class of the majority of its nearest neighbors. Whereas, K-means is a type of unsupervised learning algorithm which is extensively beneficial for clustering.

In k-means clustering, the algorithm needs a set of unlabeled points and a threshold as the base. The algorithm even takes unlabeled data and tends to learn the technique to cluster it into groups by computing the mean of the distance between different unlabeled points present in the dataset.

10. List out the different types of Algorithms in Machine Learning.

The different types of algorithms in machine learning are:

• Supervised Learning
• Semi-supervised Learning
• Unsupervised Learning
• Reinforcement Learning

11. Describe in your own terms what you understand by the Reinforcement Learning technique?

Reinforcement learning is an algorithm technique extensively used in Machine Learning. It involves the effective working of an agent that interacts with its environment by constructing actions and discovering penalties or rewards.

We can deploy Reinforcement learning with the help of different software and machines to look out for the best suitable behavior or path it should act on in a specific situation. It primarily tries to learn on the basis of rewards or penalties given to it for every action it performs.

12. How would you describe the trade-off between bias and variance?

Both bias and variance depict errors. Bias can be thought of as an error due to erroneous or overly simplistic assumptions that we may adopt in the learning algorithm. It can lead to problems like the model under-fitting the data, making it difficult to have high predictive accuracy and even making it troublesome in generalizing the knowledge from the training set to the test set.

On the other hand, we can think of Variance as an error that exists due to too much complexity present in the learning algorithm. It leads to problems such as the algorithm being highly sensitive to high degrees of variation in the training data of the algorithm, which in turn leads the model to overfit the data.

13. How would you describe the term ensemble learning?

Numerous models, for instance, classifiers, are strategically made and combined in an attempt to solve a specific computational program, and we can call this ensemble learning. We may even encounter alternative terms for the ensemble methods such as committee-based learning or learning multiple classifier systems. It tends to train various hypotheses to resolve the same issue.

One of the most opposite examples of ensemble modeling is the random forest trees where we make use of several decision trees to predict outcomes. We can extensively make use of it to enhance the classification, function approximation, prediction, and many such criteria of a model.

14. What do you understand about model selection in Machine Learning?

The process of electing models among diverse mathematical models, which we can deploy to define the same data is known as Model Selection. We apply the concept of Model learning in the fields of statistics, data mining, and machine learning.

15. Discuss the three stages of building hypotheses or models in machine learning.

The architecture for building the hypotheses or model in machine learning comprises 3 stages:

a. Model building: It elects a suitable algorithm for the model and trains it according to the requirement of the problem.

b. Applying the model: It is accountable for checking the accuracy of the model through the test data.

c. Model testing: It executes the necessary changes after testing and applies the final model.

16. Define the Training Set and Training Test in your own words.

In certain areas of information of machine learning, we make use of a set of data in an attempt to discover the potentially predictive relationship, which is known as the Training Set. The training set happens to be an example that we facilitate to the learner. B

esides, we make use of the Test Set to examine the accuracy of the hypotheses that the learner generates. It transpires to be the set of instances held back from the learner. Thus, we can infer that the training set is distinct from the test set.

17. List the common ways to handle missing data in a dataset.

Missing data is one of the prevailing factors while working with data and handling. We can consider it as one of the greatest challenges that data analysts face. There are numerous ways in which one can impute the missing values.

Some of the conventional methods to handle missing data in datasets are deleting the rows (reduction), replacing them with mean/median/mode, predicting the missing values, assigning a unique category, using algorithms that are able to handle missing values, and many more.

18. Describe the term ILP?

ILP is the acronym for Inductive Logic Programming. It is a component of machine learning which makes use of logic programming. It extensively focuses on searching patterns in data which we can use to build predictive models. In this process, we adopt an assumption that the logic programs are a hypothesis.

19. State the necessary steps involved in the Machine Learning Project.

There are quite a few essential steps we must abide by to achieve a good working model while doing a Machine Learning Project. Those steps include parameter tuning, data preparation, data collection, training the model, model evaluation, and prediction, and many others.

20. What do you understand by Precision and Recall?

Precision and Recall both are the measures that we may use in the information retrieval domain. They facilitate us to measure the extent of accuracy of an information retrieval system that reclaims the related data as requested by the user.

Precision can be thought of as a positive predictive value. It is the ratio of relevant instances among the received instances.

On the other hand, recall is the ratio of relevant instances that the algorithms have retrieved over the total amount of relevant instances. The recall is also known by the term sensitivity.

21. Describe Decision Tree in Machine Learning.

We can define Decision Trees as Supervised Machine Learning, where the algorithm continuously splits the data according to a certain parameter. It tends to build classification or regression models as similar as a tree structure, with datasets that are split up into ever-smaller subsets while developing the decision tree.

We can define the trees with the help of two entities, namely decision nodes, and leaves. The leaves of the tree are the decisions or the outcomes, and the decision nodes are where the algorithm splits the data. Decision trees are able to manage both categorical and numerical data.

22. List out some of the functions of Supervised Learning.

• Classification
• Speech Recognition
• Regression
• Predict Time Series
• Annotate Strings

23. List out some of the functions of Unsupervised Learning.

• Finding out clusters of the data
• Looking for low-dimensional representations of the data
• Recognizing interesting directions in data
• Looking out novel observations/ database cleaning
• Recognizing interesting coordinates and correlations

24. Describe algorithm-independent machine learning in your own terms.

We can define Algorithm independent machine learning as machine learning, where the usage of mathematical foundations is independent of any particular classifier or learning algorithm.

25. What do you mean by the classifier in machine learning?

A classifier is a case of a hypothesis or discrete-valued function that the algorithm uses in an attempt to assign class labels to particular data points. It is a system that takes in the input of a vector of discrete or continuous feature values and outputs a single discrete value, or the class.

26. What do you understand about Genetic Programming?

Genetic Programming (GP) is strikingly similar to an Evolutionary Algorithm, a subset of machine learning. Genetic programming software systems try to implement an algorithm that makes use of random mutation, a fitness function, crossover, and multiple generations of evolution in an attempt to resolve a user-defined task. The genetic programming model lays its basis on testing and selecting the best option among a set of results.

27. What do you understand about SVM in machine learning? What are the different classification methods that SVM can handle?

SVM is the acronym for Support Vector Machine. SVM is a type of supervised learning model with an associated learning algorithm that analyzes the data that the algorithm uses for classification and regression analysis.

The classification methods that SVM can handle are as follows:

• Combination binary classifiers
• Modification of binary to incorporate multiclass learning

28. Elucidate True Positive, True Negative, False Positive, and False Negative in Confusion Matrix with an example.

1. True Positive: When a model correctly envisages the positive class, we can call this situation to be a true positive.

For example, Umpire giving a Batsman NOT OUT when he is NOT OUT.

2. True Negative: When a model correctly envisages the negative class, we can call this situation to be a true negative.

For example, Umpire giving a Batsman OUT when he is OUT.

3. False Positive: When a model incorrectly envisages the positive class, we can call it to be a false positive. It is also known by the name ‘Type I’ error.

For example, Umpire giving a Batsman NOT OUT when he is OUT.

4. False Negative: When a model incorrectly envisages the negative class, we can call it to be a false negative. It is also known by the name ‘Type II’ error.

For example, Umpire giving a Batsman OUT when he is NOT OUT.

29. What, in your preference, is more important between model accuracy and model performance?

We can think of Model accuracy as a subset of model performance. The accuracy of the model is in direct proportion to the performance of the model. Thus, we can infer that the better the performance of the model, the more accurate are the predictions of the model.

30. What do you understand by Bagging and Boosting?

Bagging is a process in ensemble learning that the algorithm uses for enhancing unstable estimation or classification schemes.

Any algorithm makes use of Boosting methods sequentially to reduce the bias of the combined model.

31. How would you draw the similarities and differences between bagging and boosting in Machine Learning?

Similarities of Bagging and Boosting

• Both of them are ensemble methods to get N learns from 1 learner.
• Both of them generate several training data sets with random sampling.
• Both of them generate the final result by taking the average of N learners.
• Both of them reduce variance and provide higher scalability.

Differences between Bagging and Boosting

• Although we build them independently, for Bagging, Boosting tries to make the addition of new models which perform well where previous models are bound to fail.
• Only Boosting is able to determine the weight for the data to tip the scales in the favor of the most challenging cases.
• Only Boosting makes an attempt to reduce bias. However, Bagging may solve the problem of overfitting while boosting can increase it.

32. What is Cluster Sampling?

Cluster Sampling is a process of randomly choosing intact groups within a defined population, sharing similar characteristics. A Cluster sample relates to a probability where each sampling unit is a collection or cluster of elements.

33. Discuss Bayesian Networks.

Bayesian Networks more commonly referred to as belief networks or casual networks are beneficial to represent the graphical model for probability relationships among a set of variables within the dataset.

For example, we can make use of a Bayesian network to represent the probabilistic relationships between diseases and their corresponding symptoms. As per the given symptoms, the network can even compute the probabilities of the presence of various diseases associated with them.

34. What are the two components that a Bayesian logic program is fabricated from?

A Bayesian logic program is a composition of two components:

• Logical: It consists of a set of Bayesian Clauses, which tries to capture the qualitative structure of the domain.
• Quantitative: The algorithm makes use of it to encode quantitative information about the domain.

35. How do you interpret dimension reduction in machine learning?

Dimension reduction is the process that we can deploy to reduce the number of random variables under consideration within the given dataset.

We can further divide Dimension reduction into feature selection and extraction.

36. Why do we sometimes refer to an instance-based learning algorithm as a Lazy learning algorithm?

In machine learning, we can describe lazy learning can as a method where the algorithm delays the induction and generalization processes until it performs classification. Owing to the same property, an instance-based learning algorithm is sometimes referred to as a lazy learning algorithm.

37. What do you mean by the F1 score?

The F1 score attempts to represent the measurement of a model’s performance. We may refer to it as a weighted average of the precision and recall of a model. Its results tending to 1 are considered as the best, and those tending to 0 as the worst. We can even use it in classification tests, where true negatives do not have much significance.

38. How can you prune a decision tree?

We can infer that Pruning has occurred in decision trees when we remove the branches which may consist of weak predictive power in an attempt to reduce the complexity of the model and enhance the predictive accuracy of a decision tree model. Pruning may occur bottom-up and top-down, along with approaches such as reduced error pruning and cost complexity pruning.

39. What do you understand about the Recommended Systems?

Recommended System accounts to be a sub-directory of information filtering systems. It attempts to predict the preferences or rankings that a user offers to a product. According to the preferences of the user, it tends to provide similar recommendations to a user for future use. Recommendation systems find themselves extremely useful in movies, news, research articles, products, social tips, music, and many such areas.

40. What is Underfitting?

Underfitting occurs when we possess a low error in both the training set and the testing set. Few algorithms may work better for interpretations but may even fail for better predictions.

41. State the situations in which regularization becomes unavoidable in Machine Learning.

Regularization becomes necessary whenever the model begins to encounter overfit/ underfit. It represents a cost term for bringing in more features with the objective function. As a consequence of this, it attempts to push the coefficients for many variables to zero and thus reduce cost terms. It even aids to reduce the model complexity so that the model can become outdo its job at predicting (generalizing).

42. What do you understand by Regularization? What kind of problems do we solve with regularization?

A Regularization is a form of regression, which tends to constrain/ regularize or shrink the coefficient estimates towards zero. In simple words, it discards learning a more complex or flexible model in an attempt to avoid the risk of overfitting. It even reduces the variance of the model, without a substantial increase in its bias within the previous range.

We make use of Regularization to address overfitting problems as it tends to penalize the loss function by adding a multiple of an L1 (LASSO) or an L2 (Ridge) norm of weights vector w.

43. Why are we required to convert categorical variables into factors? Which functions do we use to perform the conversion?

Most Machine learning algorithms necessitate numbers as input. That is why we need to convert categorical values into factors to attain numerical values. We do not even have to deal with dummy variables.

We make use of the functions factor () and as. factor () to convert variables into factors.

44. Where can you find instances of machine learning in day-to-day life?

The majority of people are already using machine learning in their everyday life. Assume that you are engaging with the internet, where you are actually expressing your preferences over the search results, likes, dislikes through your searches. Cookies attempt to collect all these things coming on your computer, from this, it evaluates the behavior of a user. It even facilitates the progress of a user through the internet and provides similar suggestions in the future.

We can even consider the navigation system as one of the examples where we are using machine learning to estimate the distance between two places using optimization techniques and even judge their situations.

45. How will you differentiate between deep learning and machine learning?

Machine Learning comprises algorithms that tend to learn from patterns of data and then apply it to the decision-making process. Deep Learning, on the other hand, is capable of learning through processing data on its own and has striking similarities to the human brain where it identifies something, analyzes it, and makes a decision.

The key differences between these two are as follow:

• The manner in which they present data to the system.
• Machine learning algorithms always entail structured data and deep learning networks rely on layers of artificial neural networks.

46. What is the Time series?

We can define a Time series as a sequence of numerical data points in successive order. It tends to track the movement of the chosen data points over a particular period of time and records the data points at regular intervals. Time series does not entail any minimum or maximum time input. Analysts often make use of Time series to examine data according to their specific requirements from the model.

47. State the criteria through which you select important variables while working on a data set?

There are various means to select important variables from a data set that are inclusive of the following:

• Identifying and discarding correlated variables before finalizing on important variables
• We can select the variables based on ‘p’ values from Linear Regression
• Forward, Backward, and Stepwise selection
• Lasso Regression
• Random Forest and plot variable chart of the variable
• We can even consider top features based on information gained for the available set of features.

48. Differentiate covariance and correlation from one another?

Covariance tends to measure how two variables are related to each other and how one of them would vary with respect to the changes in the other variable. If the value of Covariance is positive, it indicates there is a direct relationship between the variables and, one of them would increase or decrease with an increase or decrease in the base variable respectively, provided that all other conditions remain constant.

Correlation tends to quantify the relationship between two random variables and possesses only three specific values, and those are 1, 0, and -1.

1 signifies a positive relationship, -1 represents a negative relationship, and 0 implies that the two variables are independent of each other.

49. What is the difference between causality and correlation?

We apply Causality to situations where one action, considering X, roots an outcome, say Y, whereas Correlation is just relating a particular action (X) to another distinct action(Y) however, X is not the necessary cause of Y.

50. You are given a data set about utilities fraud detection. You have built a classifier model and even achieved a performance score of 98.5%. Can you infer from this that it is a good model? If yes, justify. If not, what can you do about it to enhance the performance?

As we know, the data set about utility fraud detection is not balanced enough implying that it is imbalanced. In such a data set, we cannot hold accuracy scores accountable as the measure of performance as it may only be predicting the majority class label accurately. However, in this case, our point of interest is to attain predictive accuracy in the minority label.

But the model often treats minorities as noise and ignores them. As a consequence of this, there is a high probability of misclassification of the minority label in comparison to the majority label. For effectively evaluating the model performance in case of imbalanced data sets such as this one, we should make use of the Sensitivity (True Positive Rate) or Specificity (True Negative Rate) to evaluate the class label-wise performance of the classification model.

In case, where the minority class label performance is not up to the mark, then we could perform the following:

• We can make use of under-sampling or over-sampling to balance the data.
• We can even change the prediction threshold value.
• We can also assign weights to labels such that the minority class labels receive the larger weights.
• We could even detect anomalies.

Conclusion

We have thus reached the end of the article where we discussed the important Machine Learning Interview questions. These questions would help you ace your interviews better and even boost your knowledge about the basic concepts. Though we have not covered any questions that deal with a specific programming language, we have tried our best to curate a list of important questions regarding the overall Machine Learning topic.