Unsupervised Learning Algorithms

Unsupervised learning is a type of machine learning where the model is not given any labeled data to train on. Instead, the model must find patterns or relationships in the data on its own. Some common unsupervised learning Algorithms include

  1. Clustering: Clustering is the process of grouping similar data points together. Clustering algorithms try to find patterns in the data and group similar data points together. Common clustering algorithms include K-means, Hierarchical clustering, and DBSCAN. K-means is a popular algorithm that partitions the data into k clusters based on their similarity. Hierarchical clustering, on the other hand, creates a hierarchical tree-like structure of clusters, where each node represents a cluster. DBSCAN is a density-based clustering algorithm that groups together data points that are close to each other in the feature space.

  2. Dimensionality reduction: Dimensionality reduction is the process of reducing the number of features in the data. This is useful when the data has a high number of features, which can make the model hard to train and lead to overfitting. Common dimensionality reduction algorithms include PCA (Principal Component Analysis), t-SNE (t-Distributed Stochastic Neighbor Embedding), and Autoencoder. PCA is a linear algorithm that reduces the data to a lower-dimensional space by finding the principal components of the data. t-SNE is a non-linear algorithm that reduces the data to a 2D or 3D space while preserving the structure of the data. Autoencoder is a neural network that learns to reconstruct its input by compressing it to a lower-dimensional space and then expanding it back to the original space.

  3. Anomaly detection: Anomaly detection is the process of identifying data points that are unusual or abnormal. Anomaly detection is useful in many applications such as fraud detection, network intrusion detection, and medical diagnosis. Common anomaly detection algorithms include Isolation Forest, Local Outlier Factor, and Autoencoder. Isolation Forest is an algorithm that isolates anomalies by randomly selecting a feature and then randomly selecting a split value between the maximum and minimum values of the selected feature. Local Outlier Factor (LOF) is an algorithm that identifies outliers by measuring the local deviation of a given data point with respect to its neighbors. Autoencoder can also be used for anomaly detection by training it on normal data and then using it to detect abnormal data points by looking at the reconstruction error.

  4. Autoencoder: An autoencoder is a neural network that learns to reconstruct its input. Autoencoder consist of an encoder and a decoder. The encoder compresses the input data into a lower-dimensional representation, and the decoder expands the lower-dimensional representation back to the original data. Autoencoder can be used for dimensionality reduction and anomaly detection, as well as for generative modeling.

  5. Generative models: A generative model is a model that learns to generate new examples similar to the ones in the training data. Generative models are useful for tasks such as image synthesis, text generation, and video prediction. Some popular generative models include Generative Adversarial Networks (GANs) and Variational Autoencoder (VAE). GANs consist of two neural networks: a generator that generates new data, and a discriminator that tries to distinguish the generated data from the real data. VAEs are a type of generative model that learns a probabilistic mapping from the data to a lower-dimensional latent space and then generates new data by sampling from this latent space.

  6. Self-organizing maps: Self-organizing maps (SOMs) are a type of unsupervised neural network that can be used for dimensionality reduction, clustering, and visualization of high-dimensional data. SOMs consist of a two-dimensional grid of neurons, where each neuron represents a cluster of similar data points. The SOM algorithm trains the network by adjusting the weights of the neurons so that similar data points are mapped to nearby neurons. SOMs can be used to reduce the dimensionality of the data by mapping it to the two-dimensional grid of neurons and can also be used to visualize high-dimensional data by creating a 2D map of the data.

  7. Expectation-Maximization: Expectation-Maximization (EM) is a technique for finding the maximum likelihood estimates of the parameters of a model when some of the data is missing or hidden. EM is particularly useful for mixture models, where the data is assumed to be generated by a mixture of different probability distributions. EM iteratively estimates the parameters of the model by alternating between an expectation step, where the hidden data is estimated given the current model parameters, and a maximization step, where the model parameters are updated given the estimated hidden data.

  8. Generative topographic mapping: Generative topographic mapping (GTM) is a type of unsupervised neural network that can be used for dimensionality reduction and visualization of high-dimensional data. GTM is similar to SOMs, but it uses a probabilistic approach to map the data to a lower-dimensional space. GTM consists of two components: a generative component that models the probability density of the data, and a topographic component that maps the data to a lower-dimensional space. GTM can be used to reduce the dimensionality of the data and to visualize high-dimensional data by creating a 2D map of the data.

  9. Restricted Boltzmann Machine: A Restricted Boltzmann Machine (RBM) is a type of generative stochastic artificial neural network that can learn a probability distribution over its set of inputs. RBMs have been successfully applied to collaborative filtering, feature learning, topic modelling and even protein folding.

  10. Deep Belief Network: A Deep Belief Network (DBN) is a generative probabilistic model that is composed of multiple layers of RBMs stacked on top of each other. DBNs can be trained in an unsupervised manner by pretraining each RBM one at a time and then fine-tuning the whole network using supervised backpropagation. DBNs have been used for various tasks such as image recognition, natural language processing, and speech recognition.

  11. Hidden Markov Models: A Hidden Markov Model (HMM) is a statistical model that is often used for sequential data such as time series and speech signals. HMMs consist of a set of states and a set of observations, where the true state of the system is hidden and can only be inferred from the observations. HMMs can be trained using the Baum-Welch algorithm, which is a type of expectation-maximization algorithm that estimates the parameters of the model given the observations. HMMs have been used for various tasks such as speech recognition, natural language processing, and bioinformatics.

  12. Factor analysis: Factor analysis is a statistical technique that is used to identify the underlying factors or latent variables that explain the relationships between a set of observed variables. Factor analysis can be used to reduce the dimensionality of the data, identify patterns in the data, and simplify the data for further analysis. There are several different methods for factor analysis including principal component analysis (PCA), common factor analysis, and principal axis factoring.

  13. Independent Component Analysis: Independent Component Analysis (ICA) is a computational technique that is used to separate a multivariate signal into independent non-Gaussian components. ICA can be used to identify the underlying sources of a signal, such as speech signals, and to separate them into their individual components. ICA is based on the assumption that the sources are statistically independent and non-Gaussian.

  14. Non-Negative Matrix Factorization: Non-Negative Matrix Factorization (NMF) is an unsupervised technique that factorize a non-negative input matrix V into two non-negative matrices W and H such that V ≈ WH. NMF can be used for tasks such as topic modeling, dimensionality reduction, and feature extraction.

  15. Latent Semantic Analysis (LSA) also known as Latent Semantic Indexing (LSI) is a technique that uses Singular Value Decomposition (SVD) to reduce the dimensionality of a term-document matrix. LSA is commonly used for natural language processing tasks such as text classification, information retrieval, and text summarization. It can be used to identify the underlying themes or topics in a collection of documents, and to represent documents and queries in a lower-dimensional space.

  16. Latent Dirichlet Allocation (LDA) is a generative probabilistic model that is used to discover the latent topics present in a corpus of text. LDA is a technique for topic modeling, which is the task of identifying the underlying themes or topics in a collection of documents. LDA represents each document as a mixture of multiple topics and each topic is represented by a probability distribution over the words in the vocabulary.

  17. Non-negative Tensor Factorization (NTF) is a generalization of Non-negative Matrix Factorization (NMF) to multi-dimensional arrays. NTF can be used for tasks such as image and video analysis, text mining, and bioinformatics.

  18. Spectral Clustering: Spectral Clustering is a technique that can be used for clustering data that is not linearly separable. Spectral Clustering operates by creating a graph of the data points and then clustering the graph into different clusters. Spectral Clustering is based on the eigenvectors of the Laplacian matrix of the graph and it has been used for tasks such as image segmentation, gene clustering, and text clustering.

  19. Hierarchical density-based clustering (HDBSCAN) is a density based clustering algorithm that identifies clusters of arbitrary shape and size, and that can also identify noise points. It’s based on the concept of density reachability and density connectivity.

  20. Subspace Clustering: Subspace Clustering is a technique used for clustering data points that lie in a high-dimensional space. Subspace Clustering algorithms attempt to find clusters in the data by identifying subspaces of the data that contain clusters. Examples of subspace clustering algorithms include Principal Component Analysis (PCA)-based Clustering and Linear Discriminant Analysis (LDA)-based Clustering.


There are many other unsupervised learning techniques such as Multi-dimensional Scaling, Nearest-neighbor methods, Markov Chain, etc. Each technique has its own strengths and weaknesses, and the choice of technique depends on the specific problem and the type of data. Additionally, many techniques have multiple variations and modifications that can be used for specific tasks. It's important to note that these techniques are not mutually exclusive, many of them can be combined to leverage their strengths and overcome their limitations. Moreover, the choice of technique should be guided by the characteristics of the data and the problem at hand, and it's often useful to experiment with multiple techniques to find the one that works best.