Modelling

Today, we're diving into the Modelling phase of the AI Project Cycle. This is where we create models that help us predict outcomes based on our data.

What do you mean by 'models'? Are they like a virtual version of something?

That's a good way to think about it! Models are mathematical representations that learn from data. By exposing them to different data points, they begin to understand patterns that help predict future events. Essentially, they help us make data-driven decisions!

So, how do we start creating these models?

Great question! We start by splitting our data into two parts: the training set, which the model learns from, and the testing set, which we use to evaluate performance later.

When it comes to AI models, we mainly deal with two types: supervised and unsupervised learning. Who can tell me the difference between them?

I think supervised learning uses labeled data, while unsupervised doesn’t?

Exactly! In supervised learning, we train our models using historical data that includes the correct answers, like if an email is spam or not. In unsupervised learning, we don’t have labels, so the model tries to identify patterns by itself, like grouping similar customers.

Can you give an example of each?

Sure! An example of supervised learning could be a model that predicts housing prices based on various features, like location and size. An example of unsupervised learning might be using clustering to categorize users based on their purchasing behaviors.

After we've chosen our algorithm and split the data, our next steps are training and testing our model. Can anyone explain why we need to test our model?

So we can see if it actually works, right?

Exactly! Testing allows us to see how well our model predicts outcomes using new data it hasn't seen before. If it performs well, we can be more confident in its predictions.

But what if it performs poorly?

Good question! If the model isn’t accurate, we may need to improve the quality of our data, explore different algorithms, or fine-tune how we train it.

Would it be easier just to change the data instead of the model?

Sometimes, yes! Improving the data can be quicker and more effective than adjusting the model itself.

Let’s apply what we’ve learned! Imagine we want to predict food waste in our school canteen. First, what kind of data would we need?

We would need information like the number of students and what meals are being served.

Right! And once we have our data, how would we go about building our model?

We would split the data into training and testing, choose an algorithm, and then train it.

Correct! Once trained, we would test it to see how accurately it predicts waste. If it predicts correctly, we can help the canteen reduce food waste effectively.

And if it doesn’t predict well?

Then we review our data and model to identify improvements. The goal is always to provide actionable insights!