This content originally appeared on DEV Community and was authored by SOMYA RAWAT
KNN Algorithms :
Best Cases :
• If dimensionality (no of features) is low then this
works best.
• If you know the right distance measure then
KNN is a good option.
Worst Cases :
• If dimensionality (no of features) is low then this
works best.
• If you know the right distance measure then
KNN is a good option.
Naive Bayes Algorithm :
*Best Cases :*
• If the Conditional independence assumption of
naive Bayes is true then it performs very well.
• Naive Bayes is the default algorithm when solving
text classification problems.
• Naive Bayes is often used when you have
categorical features (binary)
• Great interpretability, feature importance, low run
time.
Worst Cases :
• If the Conditional independence assumption of
naive Bayes is false then its performance
deteriorates.
• Naive Bayes is not often used when you have real
value features.
• Easily get's overfitted when Laplace smoothing is
not done correctly.
Logistic Regression Algorithm :
Best Cases :
• It works best when data is almost linearly
separable and it is good if you have a low latency
requirement.
• It is good for interpretability and feature
importance using weights (coefficients).
• Less impact of outliers because of the sigmoid.
• If dimensionality is large it works well.
Worst Cases :
• It works badly when data is not linearly separable.
• When data is imbalanced.
• Missing values.
• No multi-class classification for the base model.
• When multicollinearity exists these models will
not work properly.
Linear Regression Algorithm :
Best Cases :
• It is good for interpretability and feature
importance using weights (coefficients).
• If Feature engineering is done then this model
can work better.
Worst Cases :
• Outliers impact a lot.
• When multicollinearity exists these models will
not work properly.
SVM Algorithm :
Best Cases :
• If you can find the right kernel then thins works at its
best.
• This can be applied to non-linear problems.
• Interpretability and feature importance is easy for
linear SVM's.
• The impact of outliers is less.
• If dimensionality is large then SVM works like a charm.
Worst Cases :
•Interpretability and feature importance is hard for
kernel SVM's.
• If training data is large, training time is high.
Decision Tree Algorithm :
Best Cases :
• Multi-class classification is possible.
• Interpretability and feature importance.
Worst Cases :
• Imbalanced data impacts a lot.
• If dimensionality is large then training time is high.
• If you use one-hot encoding then training time will
be high.
• Outliers will impact the model.
This content originally appeared on DEV Community and was authored by SOMYA RAWAT
SOMYA RAWAT | Sciencx (2022-04-19T01:43:50+00:00) Best and Worst Cases of Machine Learning Algorithms ?. Retrieved from https://www.scien.cx/2022/04/19/best-and-worst-cases-of-machine-learning-algorithms/
Please log in to upload a file.
There are no updates yet.
Click the Upload button above to add an update.