What are the Models of UnSupervised Learning

 

In Unsupervised Learning, the model understands the data without any supervision. The techniques applied in this Learning cluster unlabeled data based on similarities and differences.

This Machine Learning category follows the below working process—

1. Collect Unlabeled Data
2. Select an Algorithm
3. Train the Model on Raw Data
4. Group or Transform Data
5. Interpret and Use Results

Unsupervised Machine Learning can be broken down into following categories—

• Clustering
• Association rules
• Dimensionality reduction.

Each of these categories consists of models whose functionality we will discuss in this article

Clustering

These algorithms find the relationship patterns among data samples and then cluster those samples into groups having similarity based on features. Most of the industries use clustering models, from airlines to healthcare and beyond.

K-Means Clustering

• It is also called as Flat Clustering algorithm.
• This algorithm divides the dataset into K clusters based on feature similarity.
• The objective is to group similar data points while keeping different groups separately.
• K-Means Clustering follows the below working process:

a)     Choosing Number of Clusters (k): The first step is to decide total clusters (k) you want to create. Choosing k is essential for meaningful clustering.

b)     Initial Partitioning: After k is chosen, the model partitions the data points into k subsets.

c)     Computing Centroids: After the initial partition, the model determines the centroid for each cluster.

d)     Reassigning Points: Each data point is assigned to the cluster whose centroid is nearest, typically with the help of Euclidean distance.

e)     Iteration Until Convergence: Steps c) and d) are repeated until the cluster assignments no longer change

Hierarchical Clustering

• Groups similar data points into hierarchy of clusters, allowing us to explore data at multiple levels of granularity.
• The output is represented in the form of a tree called Dendrogram.
• Unlike K-Means, it does not require preselecting the number of clusters.
• Follows the below main approaches—

a)     Agglomerative (Bottom-Up):

>  Also called AGNES (Agglomerative Nesting)

>  This approach starts with taking all data points as single clusters and merging them until one cluster is left.

b)     Divisive (Top-Down):

>  Also called DIANA (Divisive Analysis)

>  All the data points are treated as one big cluster and the process of clustering involves dividing one big cluster into various small clusters.

DBSCAN

• DBSCAN, which stands for Density-Based Spatial Clustering of Applications with Noise is a density-based clustering algorithm that groups together data points that are closely packed while marking points in low-density regions as outliers.
• This Clustering method uses 2 important parameters—

a)     Epsilon (ε): The max distance within which two data points are considered neighbors.

b)     MinPts: The min number of neighboring points required to form a dense region (cluster).

• DBSCAN algorithm classifies the data points into 3 types—

a)     Core point: Have sufficient number of neighbors within a given radius.

b)     Border point: Close to the core point but does not meet the density requirement themselves.

c)     Noise point: Which are isolated and do not belong to any cluster.

 

Association Rules

This Learning technique finds relationships, patterns or associations between variables in large datasets. It is commonly used in Market Basket Analysis for identifying products that are most frequently purchased.

Apriori Algorithm

• It was introduced by Rakesh Agrawal and Ramakrishnan Srikant in 1994.
• Finds frequent itemsets in a transaction database and generate association rules based on those itemsets.
• This algorithm follows the principle— “If an itemset is frequent, then all its subsets must also be frequent.”
• The key metrics in this algorithm are--

a)     Support: measures how frequently an item or item-set appears in the dataset relative to the total transactions.

b)     Confidence: measures the likelihood that item Y is purchased when item X is purchased.

c)     Lift: measures how likely the two items are purchased together compared to random chance.

FP (Frequent Pattern)-Growth Algorithm

• This technique is an improvement to Apriori algorithm, since it efficiently discovers frequent itemsets in transactional databases without generating candidate itemsets.
• FP Tree in this algorithm is a tree data structure created from the transaction data while generating frequent itemsets.
• FP-Growth algorithm works in the following manner:

a)     First, it compresses the input database creating an FP-tree instance to represent frequent items.

b)     Then, it divides the compressed database into a set of conditional databases, each associated with one consistent pattern.

c)     Finally, each of these databases are mined separately.

Dimensionality Reduction 

Dimensionality reduction reduces the input features in a dataset while preserving important information. It transforms high-dimensional data into a lower-dimensional data for simpler representation.

Principal Component Analysis (PCA) 

• PCA was introduced by Karl Pearson in 1901 as a statistical method to analyze data variation and relationships.
• This technique transforms large set of correlated features into a smaller set of uncorrelated features called principal components, while retaining as much information as possible.
• The working process of this technique is as follows—

a)     Standardize the data

b)     Compute the covariance matrix

c)     Compute the eigenvectors and eigenvalues of the covariance matrix

d)     Select the principal components

e)     Project the data onto the new feature space

t-SNE

• t-SNE (T-distributed Stochastic Neighbor Embedding)  is a non-linear dimensionality reduction technique used for visualizing high-dimensional data in a lower-dimensional space mainly in 2D or 3D.
• This technique was introduced by Laurens van der Maaten and Geoffrey Hinton in 2008.
• Unlike PCA, which preserves overall variance, t-SNE focuses on preserving the local relationships between data points.

Conclusion

As industries continue to generate large amounts of unlabeled data, Unsupervised Learning becomes increasingly important across various areas such as healthcare, finance, retail, cybersecurity, and manufacturing. The models covered in this article provide the foundation for extracting actionable insights, improving decision-making, and enabling data-driven innovation without relying on manually labeled datasets.

What are the Models of Supervised Learning

 

In Supervised Learning, the Model is trained using labeled data, meaning the input data is paired with the correct output. The objective is to learn an association between input data samples and corresponding outputs after performing multiple training data instances.

The process involved in this Machine Learning category is as follows—

1. Collecting Labeled Data: Each data point includes an input (features) and the correct output (label).
2. Splitting the Data: The dataset is usually divided into a training set and a test set.
3. Training the Model: The algorithm learns from the training set by minimizing errors between predicted and actual labels.
4. Evaluating Performance: The model is tested on unseen data to measure accuracy.

There are 2 main types of Models in Supervised Learning—

    1. Classification:

• Used when the target variable is a category or class.
• Model learns from labeled data and predicts which class a new observation belongs to.

          2. Regression:

• Used when the target variable is a continuous numerical value.
• Model predicts a quantity rather than a category.

CLASSIFICATION MODELS

A. Logistic Regression

This model is primarily used for binary classification tasks that can help answer questions like—

• Is this email spam or not?
• Will this customer buy the product or not?
• Does this patient have the disease or not?

At its core, Logistic Regression predicts the probability that a given observation belongs to a particular class, and its probability score varies between 0 and 1.

Logistic Regression derives its name from the logistic/sigmoind function, which transforms a linear combination of input features into probabilities. This enables the algorithm to handle classification tasks effectively while remaining interpretable and computationally efficient.

The process workflow behind this Model is as follows—

a)     Collect training data.

b)     Compute a linear combination of input features:

c)     Apply the sigmoid function:

The sigmoid function converts any real value into a probability between 0 and 1.

d)     Make a prediction:

• If probability ≥ 0.5 → Class 1
• If probability < 0.5 → Class 0

B. Decision Tree

Decision tree is a hierarchical tree-based model that is used to classify or predict outcomes based on a set of rules. It consists of mainly 3 parts—

a) Root Node: Represents the entire dataset and the first feature used for splitting.

b) Decision Nodes: Internal nodes where the data is split based on feature values.

c) Leaf Nodes: Final nodes that represent the predicted class.

While there are various algorithms in Machine Learning, Decision Tree is used because of below reasons—

• Decision Trees usually mimic human thinking ability while making a decision, so it is easy to understand.
• The logic behind the decision tree can be easily understood because it shows a tree-like structure.

C. Random Forest

This classification model uses an ensemble of decision trees to make predictions. The algorithm was first introduced by Leo Breiman and Adele Cutler in 2001 with the key idea of creating a large number of decision trees, each of which is trained on a different subset of the data.

Below are some points that explain why Random Forest algorithm is used:

• It takes less training time than other algorithms.
• It predicts output with high accuracy, even for the large dataset it runs efficiently.
• It can also maintain accuracy when a large proportion of data is missing.

The areas where this model can be applied are as follows—

• Customer churn prediction
• Fraud detection
• Stock price prediction
• Medical diagnosis
• Image recognition

D. K-Nearest Neighbors (KNN)

KNN is a lazy learning algorithm, meaning it does not build a model during training. Instead, it stores the training data and performs computations only when making predictions.

In this model, K is just a number that tells the algorithm how many nearby points or neighbors to look at when it makes a decision.

KNN algorithm works in the following manner—

a)     Choose the value of K (the number of nearest neighbors).

b)     Calculate the distance between the new data point and all training data points.

c)     Select the K closest neighbors.

d)     For classification, assign the class with the majority vote among the neighbors.

The areas where this model is applied are—

• Recommendation Systems
• Spam Detection
• Customer Segmentation
• Speech Recognition

REGRESSION MODELS

A. Linear Regression

Defined as the statistical model, it analyzes the linear relationship between a dependent variable and a given set of independent variables. This model makes predictions for continuous/real or numeric variables such as sales, salary, age, and product price.

The linear regression model provides a sloped straight line representing the relationship between the variables. Consider the image below:

Mathematically, linear regression can be represented as:

y= a0+a1x+ ε

Here,

Y= Dependent Variable (Target Variable)

X= Independent Variable (predictor Variable)

a0= intercept of the line (Gives an additional degree of freedom)

a1 = Linear regression coefficient (scale factor to each input value).

ε = random error

The values for x and y variables are training datasets for Linear Regression model representation.

Linear Regression model can be classified as follows—

a) Simple Linear Regression:

• Uses one independent variable.
• Example: Predicting salary based on years of experience.

            b) Multiple Linear Regression:

• Uses two or more independent variables.
• Example: Predicting house prices using area, number of bedrooms, and age of the house.

B. Decision Tree Regressor

The goal of this model is to predict continuous values such as prices or scores using a tree-like structure. Unlike linear regression, decision trees partition the feature space in a hierarchical, rule-based way that enables them to capture complex, non-linear relationships.

This model continuously splits data into subsets, based on the features that result in the lowest prediction error, forming a tree-like structure where:

• Each internal node represents a decision rule on a feature.
• Each branch represents the outcome of a decision.
• Each leaf node provides the predicted value

Decision Tree Regressors use one of the following criteria for splitting the data:

• Mean Squared Error (MSE) – Measures the average squared difference between actual and predicted values.
• Mean Absolute Error (MAE) – Measures the average absolute difference between actual and predicted values.
• Friedman MSE – A variant of MSE commonly used in gradient boosting.

C. Random Forest Regressor

It is an ensemble learning method that combines multiple decision trees to produce more accurate and stable predictions. Unlike Decision Tree Regressor, which relies on a single tree, this model builds many decision trees and combines their predictions by averaging.

Random Forest Regressor works in the following manner:

1. Draw multiple bootstrap samples (random samples with replacement) from the training dataset.
2. Train a Decision Tree Regressor on each sample.
3. At each split, each tree considers only a random subset of features.
4. Each tree predicts a numerical value for a new data point.
5. The final prediction is the average of all tree predictions.

The areas where this model can be applied are:

• House price prediction
• Sales forecasting
• Demand forecasting
• Stock market prediction
• Energy consumption forecasting
• Weather prediction

CONCLUSION

The study of supervised learning demonstrates how different algorithms can be applied to solve a wide range of business and real-world problems. Classification models help organizations make informed decisions, while regression models support forecasting and trend analysis. Selecting the most suitable algorithm based on data characteristics significantly improves prediction accuracy and overall model performance.

What are the Types of Machine Learning

 

Machine Learning is a key part of Artificial Intelligence. It allows computers to learn from data and enhance their performance without needing additional programming. Machine learning algorithms identify patterns within data, which helps them make predictions, categorize information, and offer valuable insights. These abilities make machine learning useful in various fields.

There are various types of machine learning, each with unique features. In this article, we will explore all categories of machine learning.

Supervised Learning

Supervised Learning is a type of machine learning where the algorithm is trained using labeled datasets. The algorithm learns how the inputs relate to the outputs and uses that knowledge to predict results for new data. The main algorithms include:—

A. Regression:

Predicts values by finding the relationship between a dependent variable and independent variables.

B. Classification:

Predicts categorical outputs by assigning data to predefined groups.

The areas of Supervised Learning are as follows—

• Image Segmentation
• Medical Diagnosis
• Fraud Detection
• Spam detection
• Speech Recognition

Unsupervised Learning

Unsupervised Learning uses unlabeled data to train machines. The machine examines the data to uncover hidden patterns and connections. No predefined outputs are given during training. The technique organizes data based on similarities and differences.

The algorithms used in Unsupervised Learning are—

A. Clustering:

• Groups data points into clusters based on their similarity
• Helps in identifying the patterns in data without using labeled data.

B. Dimensionality Reduction: 

This reduces the number of features in a dataset while retaining important information by transforming high-dimensional data into a lower-dimensional form.

C. Association:

This method finds the relationships in a dataset. It identifies rules that show how one item is related to another.

Semi-Supervised Learning

Semi-supervised Learning combines Supervised and Unsupervised Learning. It uses two types of data: labelled and unlabeled. The labelled data contains known outputs and helps guide the model during training. The unlabeled data does not contain known outputs. 

The model identified the patterns in unlabeled dataset with the help of labelled samples. This method proves effective when only a small amount of labeled data is available.

Applications of Semi Supervised Learning include—

• Text Document Classification
• Image Recognition
• Natural Language Processing (NLP)
• Anomaly Detection

Self-Supervised Learning

Self-Supervised Learning helps the model to learn from unlabeled data by generating labels. The model learns by predicting parts of the input from other parts, creating a learning signal without the need for labeled data. As a result, an unsupervised task can be treated like a supervised task.

The areas of Self-Supervised Learning are as follows—

• Computer Vision: Enhances image-related tasks such as recognition, detection, and analysis using unlabeled data.
• NLP: Improves language understanding and tasks such as translation and sentiment analysis.
• Speech Recognition: Learn from audio data to understand speech.
• Healthcare: Supports diagnosis and analysis when labeled data is limited.
• Autonomous Systems: Helps the robots and self-driving systems to learn from sensor and video data.

Reinforcement Learning

This Learning category trains the agent to make decisions via trial and error. This Machine Learning technique follows the below workflow—

• The agent performs an action in their environment.
• Depending on the success of the action, it gets rewarded or penalized.
• After a long time, it understands the best ways to increase rewards.

Reinforcement Learning consist of two types of methods/algorithms—

A. Positive Reinforcement Learning: Specifies increasing the tendency that the required behavior would occur again by adding something.

B. Negative Reinforcement Learning: Works exactly opposite to the positive Reinforcement Learning. It increases the chance of repeating a behavior by avoiding negative outcomes.

The areas where the Reinforcement Learning is applied are as follows—

• Gaming and simulation: Teaches the agents to play and adapt.
• Robotics and automation: Enabling robots to perform tasks.
• Autonomous vehicles: Helping self-driving cars make real-time decisions.
• Healthcare and finance: Helps in optimizing treatment plans, trading and resource allocation.
• Recommendation and personalization: Improving user experience.
• Industrial and energy management: Helps in optimizing control systems and energy use.

Conclusion

This article provided an overview of the major Machine Learning categories and the types of data required by each one. The appropriate category depends on the project objective and the available data. Understanding these differences helps users choose the right approach during project development.

Process of Machine Learning Workflow

 

Machine Learning Workflow is a structured process used to build, train, evaluate, and deploy machine learning models.It takes raw data, whether structured or unstructured, and turns it into a functional model that can make accurate predictions.Having a well-defined workflow ensures efficiency, clarity, and reproducibility, while helping teams avoid common mistakes.

In this article, we are going to go through the following stages of Machine Learning process—

1. Problem Definition
2. Data Collection
3. Data Cleaning and Preprocessing
4. Exploratory Data Analysis
5. Feature Engineering and Selection
6. Model Selection
7. Model Training
8. Model Evaluation and Tuning
9. Model Deployment
10. Model Monitoring and Maintenance

Each stage in this process is important for ensuring that the model performs well in real-world conditions and doesn't fail after training.

Problem Definition

Every Machine Learning project starts with understanding what needs to be solved and how success can be measured. A clearly defined problem sets the foundation for setting project goals, expected results, and the type of solution required.

The problem can be defined by following the below Steps—

1. Identify the objective
2. Define success metrics
3. Determine constraints

Data Collection

Data is the foundation of every Machine Learning project. Collecting high quality data ensures that models can learn patterns effectively.

Data can be obtained from various sources, including:

• Internal databases
• Public datasets
• APIs
• Web scraping

Data Cleaning and Preprocessing

Before building a Machine Learning model, data must be cleaned so that it can be ready for use in an algorithm. The key preprocessing steps include—

• Handling missing values
• Removing duplicates and outliers
• Encoding categorical variables
• Feature scaling

Exploratory Data Analysis (EDA)

EDA helps uncover patterns hidden in the data by providing insights into the dataset’s structure.

This step involves the techniques as listed below—

• Descriptive statistics (Mean, Median, Variance etc.)
• Visualizations (Histogram, Scatter plot, Box plots etc.)
• Correlation analysis (Pearson, Spearman etc.)

Feature Engineering and Selection

This stage involves selecting only the most relevant features to improve model efficiency, prediction accuracy, and reduce complexity. It is where domain knowledge meets data science.

The Strategies involved in this step are—

• Polynomial features: Capture non-linear relationships.
• Interaction terms: Combine features to create new predictive signals.
• Dimensionality reduction: PCA, LDA or Feature importance ranking.

Model Selection

Selecting the correct model is important for the success of Machine Learning operations. The choice of the Model depends on below factors—

• Complexity: When choosing a model, always consider the complexity of problem and data involved.
• Decision Factors: Evaluate performance, interpretability and scalability.
• Experimentation: Try different models to find the best fit for the problem.

The most common algorithms that are involved in Model Selection are as follows

• Classification: Logistic Regression, Random Forest, SVM
• Regression: Linear Regression, Gradient Boosting
• Clustering: K-Means, DBSCAN
• Deep Learning: CNNs for images, LSTMs for sequences

Model Training

After the Model is selected, the next stage is Model Training that exposes the model to historical data allowing it to learn patterns and dependencies within the dataset.

Some of best practices in this step are—

• Split data into training, validation and test sets
• Use cross-validation to prevent overfitting
• Monitor training metrics to ensure convergence

Model Evaluation and Tuning

After the model is trained, it is crucial to evaluate its performance in real-world scenarios. Key aspects in this step are:

• Evaluation Metrics: Metrics like accuracy, precision and F1 score determine the model performance.
• Iterative Improvement: Tune the model by adjusting hyperparameters to improve predictive accuracy.
• Model Robustness: Iterative tuning helps in achieving higher levels of model robustness.

Model Deployment

Deploying a model involves turning insights into actionable results, marking the transition from experimentation to production.

The strategies that are adopted for deploying the model are—

• REST APIs (Flask, FastAPI)
• Cloud platforms (AWS SageMaker, GCP AI Platform)
• Edge devices (IoT applications)

Model Monitoring and Maintenance

To ensure that the model stays accurate over time, it needs to be continuously monitored after deployment. Regular tracking helps detect data drift, accuracy drops, or changing patterns, and retraining may be needed to keep the model reliable in real-world use.

Conclusion

Having reached the end of the article, we understood how the machine learning workflow can transform abstract concepts into actionable solutions. A well-executed workflow at every stage helps build reliable models and drive meaningful business or research impact.