DS 1

Question 1

What is the basic assumption of Naive Bayes classifier?

Answer 1

Features Are Independent

Question 2

Advantages of Naive Bayes Classifier?

Answer 2

1. Works well with many number of features (because of independence assumption)
2. Works well with large training dataset
3. Converges fast when training a model (all it needs are prior probabilities that do not change and can be stores ahead of time)
4. performs well with categorical features (use one-hot encoding. n-1 dummy variables for n categories. Assign 1 or 0 depending on whether that category is present in a row)

Question 3

Disadvantages of Naive Bayes classifier?

Answer 3

1. Correlated features affect performance

Question 4

Is feature scaling required in NBC?

Answer 4

No. Because it is based on probability and not on the distance metric

Question 5

Can NBC handle missing values?

Answer 5

It can. An attribute with a missing value is ignored when training, and if an attribute has a missing value during classifying, it is not used in the probability calculation.

Question 6

Impact of outliers in NBC

Answer 6

Robust, since we are using probabilities

Question 7

Problem statements using NBC?

Answer 7

1. Sentiment analysis
2. Document categorization
3. spam classification

Question 8

What is concept of linear regression?

Answer 8

given one or more independent features and a dependent feature, we fit a best fit line to the data. This best fit line follows the equation of the straight line y=mx+c. The best fit line is selected by minimizing a cost function (such as mean squared error). The cost function can be minimized by a gradient descent algorithm (move into the direction of the negative gradient).

Question 9

Underfitting?

Answer 9

Underfitting is a scenario in data science where a data model is unable to capture the relationship between the input and output variables accurately, generating a high error rate on both the training set and unseen data.
High bias
High Variance

Question 10

Overfitting?

Answer 10

Overfitting is a concept in data science, which occurs when a statistical model fits exactly against its training data. When this happens, the algorithm, unfortunately, cannot perform accurately against unseen data, defeating its purpose.
Low Bias
High Variance

Question 11

What is bias and variance?

Answer 11

Bias = error on training data
Variance = error on test data

Question 12

What is multicollinearity?

Answer 12

When two independent features are highly correlated (almost around 85%) and the dependent variable is also correlated with them.
Not useful to use both the features.
find which features are highly correlated using heatmap (works with smaller no. of features)
for large features, –> use ridge or lasso regression: penalizes features that are correlated

Question 13

Basic assumptions of linear regression?

Answer 13

There are four assumptions associated with a linear regression model:

1. Linearity: The relationship between X and the mean of Y is linear.
2. Homoscedasticity: The variance of residual is the same for any value of X.
3. Independence: Observations are independent of each other.
4. Normality: For any fixed value of X, Y is normally distributed.

Question 14

Advantages of linear regression?

Answer 14

1. works exceptionally well for linearly separable data
2. converges quite fast and it is easy to train
3. handle overfitting using dimensionality reduction and regularization

Question 15

Disadvantages of linear regression?

Answer 15

1. Sometimes Lot of Feature Engineering Is required
2. If the independent features are correlated it may affect performance
3. It is often quite prone to noise and overfitting

Question 16

Is feature scaling required in Linear regression?

Answer 16

Yes

Question 17

Impact of missing values and outliers in LR?

Answer 17

affected by both

Question 18

What are the advantages of SVM?

Answer 18

1. works quite well when there is a clear separation between classes
2. memory efficient
3. more effective in high dimensional spaces
4. any complex problem can be solved using SVM kernels
5. works well even with structured or semi-structured data

Question 19

What are disadvantages of SVM?

Answer 19

1. not suitable for large training datasets
2. target classes overlapping
3. SVM underperforms when no. of features exceeds no. of training samples

Question 20

Is feature scaling required for SVM?

Answer 20

Yes

Question 21

Is SVM sensitive to missing values?

Answer 21

yes. No easy accomodation of missing covariate information

Question 22

Is SVM sensitive to outliers?

Answer 22

yes, presence of even a few outliers can lead to serious missclassification

Question 23

How do we solve underfitting and overfitting in SVM?

Answer 23

we use a soft margin instead of a hard one (allow some points inside the margins) though these are still penalized. This makes sure our model doesn’t overfit.

Question 24

SVM use cases?

Answer 24

Text and hypertext categorization

handwriting recognition

Question 25

What are the basic assumptions of logistic regression?

Answer 25

a linear relationship between independent features and log odds

Question 26

What are advantages of logistic regression?

Answer 26

1. easy to understand, implement and train
2. no assumptions about distributions of classes in feature space
3. good accuracy for many simple datasets and performs well when the dataset is linearly separable
4. less inclined to overfit
5. can be extended to multiclass problems

Question 27

What are the disadvantages of logistic regression?

Answer 27

1. Sometimes lot of feature engineering is required
2. if independent features are correlated, it may affect performance
3. quite prone to noise and overfitting
4. number of observations less than number of features, logistic regression should not be used
5. non-linear problems cannot be solved because has a linear decision surface

Question 28

Is feature scaling required in Logistic regression?

Answer 28

yes

Question 29

Is logistic regression sensitive to missing values and outliers?

Answer 29

Missing values: yes

Outliers: not much, using a sigmoid function tapers the outliers