Assumptions of Machine Learning Algorithms
The performance of a machine learning algorithm on a particular dataset often depends on whether the features of the dataset satisfies the assumptions of that machine learning algorithm. Not all machine learning algorithms have assumptions this is why all algorithms differ from each other. So, in this article, I will take you through the assumptions of machine learning algorithms. If you know the assumptions of some commonly used machine learning models, you will easily learn how to select the best algorithm to use on a particular problem. Not all machine learning algorithms have assumptions this is why all algorithms differ from each other. So in the section below, I will introduce you to the assumptions of some commonly used machine learning algorithms.
Assumptions of Linear Regression:
- There is a linear relationship between dependent and independent features.
- All the features are multivariate normally.
- There is very little or no multicollinearity in the dataset.
- There is very little or no autocorrelation in the dataset.
- It also assumes that there is homoscedasticity in the data set.
Assumptions of Logistic Regression:
- It assumes that there is an appropriate structure of the output label.
- All observations are independent of each other.
- There is little or no multicollinearity in the dataset.
- It also assumes that the dataset consists of a very large sample.
Assumptions of Naive Bayes:
The Naive Bayes algorithm is said to be naive because of its naive assumption which implies that the conditional independence of causes. Simply put, the presence of one cause is not normally independent of the presence of other causes. This can be considered very difficult to accept in many cases where the probability of a particular feature is strictly correlated with another feature.
Assumptions of Support Vector Machines:
Below are the assumptions of support vector machines that you should know:
- Support Vector Machines is a family of algorithms that can operate in linear and non-linear data sets.
- Along with neural networks, SVMs are probably the best choice among many tasks where it is not easy to find a good separation hyperplane.
- Support vectors are the most useful data points because they are the most likely to be misclassified.
Assumptions of Random Forest
- No formal distributional
- random forests are non-parametric and can thus handle skewed and multi-modal data as well as categorical data that are ordinal or non-ordinal.