Deloitte Interview :
Candidate Name: Wanted to remain anonymous
ROUND 1 :
– Started with Classification particularly Imbalance , oversampling.
Which class should i oversample etc.
Telecom Churn Case Study Questions like Evaluation metric for imbalance data
what threshold to choose to diving the classes (0.5 in case of balanced else sensitivity / Specifivity etc.
What if i don’t use SMOTE() for handling imbalance how should i select the threshold now (messed up by me, roc , auc etc) Ans = Presion – Recall Curve
– NLP Questions
Sentiment analysis, preprocessing like (TFID, BOW), Embeddings, stemming, Lemmatization
libraries in know : nltk, spacy
– Regression Preprocessing
answered outlier, missing value imputation, Distribution, dummies, multicolinearity etc
You have two highy co-related columns which one will you drop? : “Based on Business Problem i will see accordingly.”, Answered
– Naive Bayes Explanation , Drawback of Naive Bayes(couln’t answer drawback of Naive Bayes, ‘Assume all are independent’, him)
– Hand Gesture Recognition Techniques (End to End)
– Resource Timesheet Forecasting . (What is it?? what you do on this?, ” Explained with a story based on what i do in TCS”.
– Do you know any Boosting Algorithms : YES
where have you used?? in Telecom Churn and Healthcare Analytics by AV
– Gradient Descent (How it works)
– KNN related. How do we choose value of K ??
– Satirical Computing:
Type 1 and Type 2 error
Alternate name of Type 1 error (couldn’t answer alternate name of Type 1 error, ‘False +ive, him)
What is p-Value (Explained with the example of Linear Regression from statsmodel)
– Do you have exposure of TimeSeires analysis : NO (didn’t ask anything and seems fine with him)
– Exp in DS/Analytics in TCS etc.
** Noted down my total exp and relevant exp **