Book My Show Interview Question | Algorithm Selection
Question
You will be aware of several algorithms. But how will you decide which machine learning algorithm to choose for your classifications problem?
in progress
1
Machine Learning
55 years
1 Answer
721 views
Great Grand Master 0
Answer ( 1 )
Support Vector Machines
– when Feature to row ratio is very high
– There are lots of outliers in data – as it considers the points very close to the classifier,
outliers are kind of ignored
Random Forest and Decision Trees
– when you are interested in significance of predictors
– You need a quick benchmark model.
– If you have messy data such as missing values, you don’t need to perform one-hot encoding,
Random forests handles missing values by itself.
Multi-layer Perceptron
– When performance is the only thing that matters.
– When control over the training process is very important.
Logistic Regression
– When impact of each predictor on the target variable is required.
– When interpretability is more important than accuracy.
XGBoost
– Probably one of the best performing models currently.
– when datasets are huge and training needs to be done quickly, parallel construction
of trees helps in reducing the training time significantly.
– Several hyperparametres can be tuned to get the best results out of the model.