Which Machine Learning Model To Use

Which machine learning model to employ?

State your problem and follow this article to know which model to use.

— if you don't know what is an ML model, have a look at this article.

Taking machine learning courses and reading manufactures about information technology doesn't necessarily tell you which auto learning model to utilise. They just requite you an intuition on how these models piece of work which may go out you in the hassle of choosing the suitable model for your trouble.

At the beginning of my journeying with ML, on solving a problem, I would try many ML models and use what works best, and I all the same do that now but I follow some best practices — on how to choose a automobile learning model — that I learned from experience, intuition and colleagues, these best practices brand things easier, hither is what I had collected.

I'll tell you lot which automobile learning model to utilise according to the nature of your problem, and I'll try to explain some concepts.

Nomenclature

First, if you have a classification trouble "which is predicting the class of a given input".

Continue in mind how many classes you'll classify your inputs to, as some of the classifiers don't back up multiclass prediction, they only support 2 class prediction.

- Wearisome merely authentic

Not-linear SVM Await at the Notation at the cease of classification'southward department for more than information about the use of SVM.
Random Woods
Neural Network (needs a lot of data points)
Gradient Boosting Tree (similar to Random Wood, but easier to overfit)

- Fast

Explainable models: Determination Tree and Logistic Regression
Non-explainable Models: Linear SVM and Naive Bayes

Annotation: SVM kernel uses (From Andrew NG'south course)

Employ the linear kernel when the number of features is larger than the number of observations.
Apply the Gaussian kernel when the number of observations is larger than the number of features.
If the number of observations is larger than 50k, speed could exist an consequence when using the Gaussian kernel; hence, one might want to apply the linear kernel.

Clustering

If you accept a clustering problem "which is dividing the data into k groups according to their features such that objects in the same group take some degree of similarity".

Hierarchical clustering (too called hierarchical cluster analysis or HCA) is a method of cluster assay which seeks to build a bureaucracy of clusters. strategies for hierarchical clustering generally fall into two types:

Agglomerative: This is a "bottom-up" approach: each observation starts in its own cluster, and pairs of clusters are merged equally ane moves upwardly the hierarchy.
Divisive: This is a "acme-down" approach: all observations start in one cluster, and splits are performed recursively equally one moves downwardly the hierarchy.

Nonhierarchical Clustering:

DBSCAN (you don't need to specify the value of the k, which is the number of clusters)
thousand-means
Gaussian Mixture Model

In instance yous are clustering a categorical information use

k-modes

Dimensionality reduction

Use The Principal Component Analysis (PCA)

PCA can be thought of as plumbing equipment an due north-dimensional ellipsoid to the data, where each axis of the ellipsoid represents a principal component. If some axis of the ellipsoid is small, and then the variance along that axis is also minor, and by omitting that axis and its corresponding main component from our representation of the dataset, we lose only a commensurately small amount of information.

In case you want to make topic modeling (explanation below) you lot utilise Singular Value Decomposition (SVD) or Latent Dirichlet Analysis (LDA), and utilise LDA in case of probabilistic topic modeling.

Topic modeling is a type of statistical model for discovering the abstract "topics" that occur in a collection of documents. Topic modeling is a frequently used text-mining tool for the discovery of hidden semantic structures in a text torso.

I hope things are easier for y'all at present, I'll update the commodity with the information I get from your feedback and your experiments.

I'll leave y'all with these 2 awesome summaries.