# How does clustering work in Machine Learning?

Machine Learning development is undeniably dependent on the quality of datasets that are incorporated into it during training sessions. The concept of Machine Learning clustering refers to a grouping of unlabeled examples. The purpose of clustering in Machine Learning is to determine the groups of similar objects with more than 2 variable quantities within the dataset. It is an unsupervised method that works in datasets where there is no target variable, nor determined relation between unlabeled data. Because it works with unlabeled data, data training isnโt an option in this processing. Some of the commonly used clustering methods in Machine Learning are:

• Density-based Clustering – dividing the data is based on the belonging probability of one dataset to a particular distribution. The grouping in this method is done based on assumptions. A commonly used assumption is Gaussian distribution.
• Hierarchical Clustering – the divided data creates a tree-like structure called a Dendrogram; which is also what itโs called when clustering trees.
• Partitioning – a.k.a centroid-based method, data is divided into non-hierarchical groups.

It is also important to mention some of the popular clustering algorithms:

• Affinity Propagation – This algorithm’s functioning is based on messages exchanged between pairs of data points until it convergences. What differenciates it from other algorithms is that it doesn’t require a specified number of clusters.
• K-Means – This algorithm requires a specified number of clusters and it classifies the dataset by dividing samples into equal variances.
• Density-based Spatial Clustering of Applications with Noise (DBSCAN) – This algorithm divides the data according to their density, by separating the areas with low and high density.

If you want to cluster your data, these are basic steps to do it properly:

1. Prepare the data – Gather the data and filter it according to your desired data outcome.
2. Create the similarity metric – Try to be as precise as possible since it will affect the result, regardless of the method chosen.
3. Run clustering algorithm – See the list above and find one that is acceptable for you, or research the other 100+ clustering algorithms available online.

Testing. CI/CD. Monitoring.