One of the most common questions that arise in machine learning problems is that, which algorithm you should use to model a given set of data. Most of the time, there is no straight forward answer, as it depends on several factors of the dataset such as the type and properties of the features, size of the data, as well as the objective of fitting the model etc. Following diagram provides a general guideline on how to select a suitable algorithm for your dataset, in WSO2 Machine Learner.
Use the above guidelines to find an algorithm to build a model to fit your data. However, if you need to find the best model, then try out a few algorithms in the same class and see which one performs better.
Algorithm types
The following table explains the algorithm types in the above diagram.
| Algorithm Type | Description | Supported Measures |
|---|---|---|
| Numerical Prediction | This involves making a numerical prediction based on the dataset analysed. | |
| Multi-class classification | This involves classifying the items in a dataset into multiple categories. | |
| Binary classification | This involves classifying the data items in a dataset into two categories. | |
| Clustering | This involves clustering the items in a dataset. | No evaluation measures are currently available. |
| Anomaly Detection | This involves identifying data items that do not confirm to the expected pattern compared to the other data items in the dataset. | |
| Deep Learning | This involves classifying data with a neural network with multiple levels, corresponding to different levels of abstraction. | |
| Recommendation | This involves using collaborative filtering techniques to predict missing entries in a dataset. |