← Back to portfolio
AIML-501 Portfolio Artifact · Workshop 1

Machine Learning Algorithm Types:
A Visual Framework

Fourteen widely used algorithms, sorted by how they learn, the kind of data they work on, and what people use them for in the real world. The lower panels show how the algorithms connect and where to start.

Bindu Bhargava Reddy Chintam
M.S. student, Indiana Wesleyan University
AIML-501: Model Development
October 2026
How to read this framework
Supervised. Learns from examples that already carry the right answer (a label).
Unsupervised. Gets no labels and finds structure, such as groups or main directions, by itself.
Unsupervised: self-supervised and generative. No human labels; the data itself supplies the target, such as the next word or a hidden word, or the model learns to create new data.
Application domains
Tabular data Computer vision NLP Generative AI Dashed = also used
Where each algorithm sits in the AI family
Artificial
intelligence
Any system that performs tasks that need human-like intelligence
Machine learning
Learns patterns from data instead of hand-written rules.

Classic ML here: Linear and Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, SVM, KNN, K-means, PCA
Deep learning
Neural networks with many layers.

Here: CNN, BERT
Generative AI
Deep models that create new text, images, or audio.

Here: GPT, GAN, Diffusion models
Supervised

Learns from labeled examples

Eight algorithms that predict a number or a category

01

Linear Regression

Linear model
Supervised
Tabular
How it works
Draws the straight line (or flat surface) that best fits past data points, then reads new predictions off that line.
Real-world use
Forecasting monthly sales; estimating a house price from its size and location.
02

Logistic Regression

Linear model
Supervised
TabularNLP
How it works
Turns a weighted sum of the inputs into a probability between 0 and 1, then picks a class when it passes a cutoff.
Real-world use
Credit default risk scores; simple spam filters built on word counts.
03

Decision Tree

Tree model
Supervised
Tabular
How it works
Asks a chain of yes or no questions about the features and follows the answers down to a label. Easy to read and explain.
Real-world use
Loan approval rules; sorting fruit photos by color share (my Workshop 1.4 project).
04

Random Forest

Tree ensemble (bagging)
Supervised
Tabular
How it works
Grows hundreds of trees, each on a random slice of rows and features, and lets them vote. The vote cancels out single-tree mistakes.
Real-world use
Credit card fraud detection; ranking which genes matter most in a disease study.
05

Gradient Boosting

Tree ensemble (boosting): XGBoost, LightGBM
Supervised
Tabular
How it works
Builds small trees one after another. Each new tree focuses on the errors the earlier trees left behind.
Real-world use
Ad click and churn prediction; a frequent winner of tabular data competitions.
06

Support Vector Machine

Margin-based model
Supervised
TabularNLPVision
How it works
Finds the boundary with the widest gap between classes. A kernel lets that boundary curve when a straight line will not work.
Real-world use
Sorting news articles by topic; recognizing handwritten digits on small datasets.
07

K-Nearest Neighbors

Distance-based model
Supervised
TabularVision
How it works
Keeps every training example and labels a new one by a majority vote of its k closest neighbors. No real training step.
Real-world use
"Customers like you also bought" suggestions; finding look-alike images.
08

Convolutional Neural Network

Deep learning (CNN)
Supervised
VisionGenAI
How it works
Slides small filters across an image to spot edges, then shapes, then whole objects, one layer at a time.
Real-world use
Face unlock on phones; flagging chest X-rays for review; reading road signs.
Unsupervised

Finds structure without labels

Classic methods on the left; self-supervised and generative deep models on the right

Classic unsupervised
09

K-means Clustering

Clustering, distance-based
Unsupervised
TabularVision
How it works
Places k center points, assigns each item to the nearest center, moves each center to the middle of its group, and repeats until nothing changes.
Real-world use
Customer segmentation for marketing; shrinking a photo down to a few colors.
10

Principal Component Analysis

Dimensionality reduction (PCA)
Unsupervised
TabularVision
How it works
Rotates the data to find the directions with the most variation and keeps only the top few, so many columns become a handful.
Real-world use
Turning thousands of gene measurements into a 2-D plot; speeding up other models.
Self-supervised and generative (deep learning)
11

BERT

Transformer encoder
Self-supervised
NLP
How it works
Reads a sentence in both directions at once. It learns by guessing words that were hidden, then gets fine-tuned with labels for a task.
Real-world use
Helping search engines understand queries; sentiment analysis of reviews.
12

GPT

Transformer decoder (LLM)
Self-supervised
NLPGenAI
How it works
Predicts the next word piece again and again, writing text one token at a time. Later tuning with human feedback shapes its answers.
Real-world use
Chat assistants such as ChatGPT; drafting code, emails, and summaries.
13

Generative Adversarial Network

Generative model (GAN)
Generative
VisionGenAI
How it works
A generator makes fake samples while a discriminator tries to catch them. Both improve until the fakes look real.
Real-world use
Making synthetic training images; sharpening low-resolution photos.
14

Diffusion Model

Generative model (e.g., Stable Diffusion)
Generative
VisionGenAI
How it works
Learns to remove noise from images step by step, so it can start from pure noise and build a brand-new picture that matches a text prompt.
Real-world use
Text-to-image tools; quick concept art and product mockups.
Domain map: where each algorithm is used
AlgorithmTabularVisionNLPGenAI
01 Linear Regression
02 Logistic Regression
03 Decision Tree
04 Random Forest
05 Gradient Boosting
06 Support Vector Machine
07 K-Nearest Neighbors
08 CNN
09 K-means Clustering
10 PCA
11 BERT
12 GPT
13 GAN
14 Diffusion Model
Filled dot = main domain. Ring = also used there. The pattern is clear: classic ML owns tabular data, CNNs own vision, transformers own language, and generative AI sits where deep vision and language models meet.
Family map: how the algorithms build on each other
LINEAR TREES DISTANCE AND MARGIN NEURAL NETWORKS Linear Regression Logistic Regression same weightedsum, plus aprobability curve Decision Tree Random Forest Gradient Boosting bagging:many trees vote boosting:each new treefixes errors K-Nearest Neighbors K-means Clustering Support Vector Machine PCA both use distance;only KNN uses labels PCA is often aprep step beforeclustering widest-gap boundary, labels needed Neural network CNNfor images Transformerfor language BERTencoder GPTdecoder understand vs. generate GAN Diffusion Model often builtfrom CNN layers denoiser is a CNN(U-Net) or a transformer builds directly on related idea or used together
Quick guide: picking a first algorithm to try
Do you have labelsfor the outcome? YES NO What kind of data?(supervised) What is the goal?(unsupervised) Tables: Gradient Boosting, or a Decision Tree to explain Images: CNN, ideally a pre-trained one Text to classify: BERT (fine-tuned) Find natural groups: K-means Too many columns: PCA first New content: GPT for text; Diffusion or GAN for images Why this order works 1. Labels decide the learning style first. 2. The data type then narrows the family: trees for tables, CNNs for images, transformers for language. 3. Start simple, measure, then move up. A clear tree beats a black box that no one can check, unless accuracy truly needs it.