Which model do I need?
Two questions narrow it down: what does your data look like, and what do you want to do with it? Find your data below, then the goal that sounds most like yours.
Rows in a spreadsheet or database
Customers, transactions, applications, products…
- Sort each record (customer, transaction, application) into a category, like yes/no or high/low risk That’s tabular classification. Plan yours · Guide: Predict customer churn from your order history
- Predict a single number for each item, like a price, a cost or a duration That’s number prediction (regression). Plan yours · Guide: Predict delivery times
- Spot rare events that look different from normal, like failing machines or suspicious activity That’s anomaly detection. Plan yours · Guide: Spot failing machines from sensor data
Numbers recorded over time
Daily sales, hourly sensor readings, website traffic…
- Predict how a number will change over time, like sales next month or demand per day That’s time-series forecasting. Plan yours · Guide: Forecast daily sales for each store
- Spot rare events that look different from normal, like failing machines or suspicious activity That’s anomaly detection. Plan yours · Guide: Spot failing machines from sensor data
Text
Emails, reviews, tickets, social posts…
- Sort pieces of text (emails, reviews, tickets, posts) into categories That’s text classification. Plan yours · Guide: Build a spam filter, start to finish
- Answer questions or write text using your own documents, like a chatbot over your manuals That’s LLM assistant (RAG chatbot). Plan yours · Guide: Build a chatbot that answers questions from your PDFs
Documents to ask questions about
PDFs, manuals, policies, a wiki…
- Answer questions or write text using your own documents, like a chatbot over your manuals That’s LLM assistant (RAG chatbot). Plan yours · Guide: Build a chatbot that answers questions from your PDFs
Images or photos
Product photos, scans, X-rays…
- Give each whole image one label, like defect/OK, a species or a diagnosis That’s image classification. Plan yours · Guide: Find defective products in photos
- Find where things are in images or video, and count or track them That’s object detection. Plan yours
Video or camera feeds
CCTV, drone footage, webcams…
- Find where things are in images or video, and count or track them That’s object detection. Plan yours
Audio or voice
Calls, meetings, voice commands, sounds…
- Work with spoken audio or sounds: transcribe, caption or recognize them That’s speech and audio. Plan yours
What people click, buy or watch
Purchase history, views, ratings…
- Suggest the right items (products, content, jobs) to each person That’s recommendation system. Plan yours · Guide: Recommend products your customers will like
Easily confused
- Classification or number prediction?
- If the answer is a category (will this customer leave: yes or no), it’s classification. If it’s an amount (how much will they spend next year), it’s number prediction. See tabular classification and number prediction (regression).
- Number prediction or forecasting?
- Forecasting predicts the next values of one series over time, mostly from its own history: sales per day next month. Number prediction estimates a value for each item from its details: the price of a house from its size and location. If the order in time is the main clue, it’s forecasting. See number prediction (regression) and time-series forecasting.
- Image classification or object detection?
- Image classification gives the whole photo one label, such as defective or OK. Object detection draws a box around each thing it finds, so you can count them or see where they are. See image classification and object detection.
- Classification or anomaly detection?
- If you have plenty of labeled examples of the bad cases, classification usually works better. If bad cases are rare, varied or new, learn what normal looks like and flag what isn’t: that’s anomaly detection. See tabular classification and anomaly detection.
- A chatbot or text classification?
- If you want answers or written text, use an LLM assistant over your documents (RAG). If you want each text sorted into fixed categories, use text classification. A language model can also sort text with no training, which makes a quick first version. See LLM assistant (RAG chatbot) and text classification.