Most Commonly Asked ML Interview Questions

Most-Commonly-Asked-ML-Interview-Questions

While Data is the new currency of the business and the industrial world, Data Science is the pathway to the next Industrial Revolution. The rising importance of data is creating a demand for skilled professionals who are well-versed in data science technologies such as Machine Learning (ML) and Artificial Intelligence (AI). However, bagging a job in the field isn’t a walk in the park. You need to be prepared to face the challenging interview process during which your mastery over a variety of data skills will be assessed such as your basic knowledge about data science and ML concepts; your ability to analyze and visualize data; your technical and programming skills, and so on.

We know that interviews can be tricky and overwhelming, and hence, we’ve prepared a list of ten most commonly asked machine learning interviews questions.

  1. What is the difference between supervised and unsupervised machine learning?

The primary difference between supervised and unsupervised learning is that while supervised learning focuses on training labeled data unsupervised learning does not require the data to be trained explicitly. For instance, for the classification function of supervised learning, one first needs to label the data that will be used to train the data model to classify the data into labeled subsets. This kind of specialized training is not required in unsupervised learning.


  1. What are parametric and non-parametric models?

Parametric models refer to those models that contain a finite number of parameters. In such a model, one only needs to know the parameters of the model to be able to predict new data. Linear regression, logistic regression, Naive Bayes, and Perceptron, are some examples of parametric models. Non-parametric models, on the other hand, contain an unlimited number of parameters and hence, are more flexible. In this model, apart from knowing the parameters of the model you also need to be aware of the state of the observed data. Decision trees, SVMs, and k-nearest neighbors are examples of non-parametric models.

  1. Explain the bias-variance tradeoff.

Predictive models usually have a tradeoff between bias and variance. While bias refers to the error occurring due to erroneous or overly simplistic assumptions in the learning algorithm being used, the variance is the error occurring due to excessively complicated assumptions in the learning algorithm in question. The purpose of the bias-variance is to  minimize the learning error of a specific algorithm by adding the bias and the variance along with some other irreducible errors due that originate from the noise in underlying datasets. For instance, you can reduce the bias by adding more variables to the model to make it complex, but in the process, you’ll add some variance to the model. Thus, to strike a perfect balance in the model, you need to have a tradeoff between bias and variance.

  1. What is the difference between stochastic gradient descent (SGD) and gradient descent (GD)?

Both SGD and GD algorithms are techniques of finding a set of parameters that can reduce the loss function of a model. The parameters are first evaluated against the data, and then adjustments are made accordingly. However, there lies a subtle difference in the approach of the two algorithms. While in GD one needs to evaluate all the training samples for each set of parameters, in SGD you need to evaluate only one training sample for the given set of parameters. Also, GD is ideal for small datasets while SGD is ideal for more massive datasets.

  1. What is the purpose of the Box-Cox transformation?

The Box-Cox transformation is a standard power transformation process of transforming datasets to facilitate normal distribution. In other words, it is used to stabilize the variance in datasets. Since most well-known statistical methods sync well with normally distributed data, it is wise to normalize the distribution using this method.

  1. Why is Naive Bayes ‘naive’?

Naive Bayes is considered to be ‘naive’ mainly because it makes such assumptions that are nearly impossible to observe in real-life data. This algorithm assumes that the presence or absence of a particular feature of a class is unrelated to the presence or absence of any other feature of the class variable in question. This entails the “absolute independence of features,” a condition that can never be fulfilled in reality.

  1. What is the difference between machine learning and deep learning?

Deep learning is a branch of machine learning exclusively concerned with neural networks. It focuses on the ways to leverage certain principles of neuroscience to model large sets of unstructured or semi-structured data with increased accuracy. To be precise, deep learning is much like an unsupervised learning algorithm that aims to ‘learn’ data representations by leveraging neural nets.

  1. How will you choose a classifier based on a training set?

In case the training set is a small, models with high bias/variance, for example, Naive Bayes, are the best fit since they are less likely to overfit. Whereas if the training set is too large, models with low bias/variance such as logistic regression are best as they can detect more complex relations in data models.

  1. What is Latent Dirichlet Allocation (LDA)?

Latent Dirichlet Allocation (LDA) is a generative model that represents documents as an amalgamation of topics, each of which has their distinct probability distribution of possible words. In other words, LDA is a technique of classifying topics or documents according to the subject matter.

  1. What is the ROC curve? What is AUROC?

The ROC (Receiving Operating Characteristic) curve is a graphical representation of the contrast between true positive rates and the false positive rate at varying thresholds. It’s mostly used to assess the sensitivity of the true positives against the false positives’ probability to trigger a false alarm.

AUROC (Area Under the Receiving Operating Characteristic) denotes a standard performance metric used to evaluate binary classification models.

Are you eager to master machine learning? Do you wish to begin a career in machine learning? Come, join our Machine Learning course at Coding Ninjas today!

Exit mobile version