Machine Learning Question Answers

BIJOY ROY Oct 30 2021 · 4 min read
Share this

1.  How you can define Machine Learning?

A:  Machine Learning is a technique using which we can help our system learn any specific task with experience.

2.  What do you understand Labelled training datasets?

A:  Labelled training dataset has a set of data that has 1 or more features and has a fixed output for those features. In other words, has both an independent variable(feature) and a dependent variable(output).

3.  What are 2 most common supervised ML tasks you have performed so far?

A:  Two common supervised ML tasks are

                                   1. Regression

                                   2. Classification

4.  What kind of Machine learning algorithm would you used to walk robot in various unknown area?

A: Reinforcement Learning

5.  What kind of ML algorithm you can use to segment your user into multiple groups?

A:   Unsupervised Learning(Clustering algo)

6.  What type of learning algorithm is realized on similarity measure to make a prediction?

A:  KNN(K Nearest Neighbour) Algorithm

7.  What is out of bag evaluation?

A:  In the case of Random Forest while doing a bootstrapping bunch of data don’t get selected. Those data are called out of bag samples. These samples can be used to evaluate the model.

8.  What do you understand by hard & soft voting classifier?

A:   In a hard voting classifier we use mode that means the class that has been predicted the most times gets selected.

         IN a soft voting classifier we use the probability of occurrence of the prediction to select the final prediction

9.  Let’s Suppose I have trained 5 different models with the same training dataset & all of them have achieved 95% precision. Is there any chance that you can combine all these models to get a better result? If yes, How? If no, Why?

A:  Yes, they can be combined. We can use a Stacking or voting classifier for this purpose.

10.  Can you please explain the difference between regression & classification?

A:  In regression, we predict continuous value but in classification, we predict discrete data.

11.  If I give you 2 columns of any dataset, what will be the steps will be involved to check the relationship between those 2 columns?

A: A correlation matrix can be used to check the relation between them and we can also plot them.

12.  Can you please explain 5 different kinds of strategies at least to handle missing values in a dataset?


  • If the data is huge then we can probably drop them.
  • We can use the mean of the data to replace
  • We can use mode to replace them
  • We can use knn imputer
  • We can use median to replace them
  • 13.  What kind of diff. issues you have faced wrt your raw data? At least mention 3 issues.


  • The data is unstructured
  • Unreliable data
  • Unknown weird symbols in data
  • 14. What is your strategy to handle categorical dataset? Explain with example


  • Label encoding: converting data to ordered numbers
  • Eg: 

    Low                                 1

    Medium                         2

    High                                3

  • One hot encoding: converting data into matrix of 0 and 1 and represent each data with a combination of 0 and 1
  • Eg:

    India                0    0    1

    Japan             0    1    0

    USA                  1   0    0

    15.  How do you define a model in terms of machine learning or in your own word?

    A: A model is a algorithm with it’s parameters tuned to represent the relation of a certain type of data

    16.  What do you understand by k fold validation & in what situation you have used k fold cross-validation?

    A:  In k fold validation we divide the data into k parts then use k-1 part for training and the remaining part for testing and we repeat this process but we switch the last part each time.

    It can be used to check for the overfitting of the model.

    17.  What is the meaning of bootstrap sampling? explain to me in your own word.

    A:  Bootstrap sampling is when we sample random samples from a dataset and create multiple mini datasets.

    18.  What is the difference between cross-validation and bootstrapping?

    A:  in bootstrapping we use sampling with replacement but in cross-validation we do sampling without replacement

    19.  Explain me the complete approach to evaluate your regression model

    A: The regression model is evaluated using several metrics like MSE score, R2 score, adjusted R2 score.

    20. Give me an example of a lazy learner and eager learner algorithms example.


    Eager learner learns the relation from the data and gives us prediction.

    Eg: Linear Regression

    Lazy learner doesn’t learn any relation but stores the dataset itself and gives prediction.

    Eg: KNN

    21. What do you understand by the holdout method?

    A: Holdout method we divide the data into train and test samples.

    22.  How do you define some features are not important for ML model? What strategy will you follow?

    A:  We can use p value, correlation matrix, feature importance from random forest.

    23.   If your model is overfitted, what you will do next?


  • Add more data
  • Add regularization
  • Do feature engineering
  • Use a simpler model
  • Tune parameters of the current model
  • 24. What is difference between correlation and covariance?A

    A:  Covariance tells us the relation between two variables and how they mode with respect to each other but it doesn’t tells us by what magnitude the movement happens

    Correlation tells us not only the movement of two variables with respect to each other but also the magnitude if movement. It ranges between -1 to 1 so 0 means no correlation closer to one is high positive correlation and closer to -1 means strong negative correlation.

    25.  What is the adv & disadvantage of naïve Bayes classifier, explain


    Adv: It is easy to understand

    GIves us probability as output

    Disadv: It expects the features to be completely independent. But that doesn’t happens in real life scenarios

    26.  Give me a situation where I will be able to use SVM instead of Logistic regression.A

    A:  When the data is complex and can’t be separated by a line. That’s when SVM can be used as it has kernel tricks using which it’ll take the data to a higher dimension and divide it using a hyperplane.

    27.  What do you understand by leaf node in decision tree?

    A:  It is the node where the output of the tree is stored

    28.  What is information gain & Entropy in decision tree?

    A:  Entropy is randomness or uncertainty of the node

    Information gain computes the information achieved before and after splitting of a node.

    Read next