Why classic machine learning first
When people hear "AI" they picture chatbots and image generators. But much of the machine learning that companies actually use runs on ordinary tables of data: credit scoring from mobile money histories, predicting which customers might leave a telecom network, estimating crop yields, flagging unusual transactions. Data science adverts in Accra describe exactly this kind of predictive modelling. Classic models are faster to train, easier to explain and often just as accurate on tabular data as deep learning.
Supervised learning
Most classic machine learning is supervised: you have past examples where you know the answer, and the model learns to predict that answer for new cases. The answer is called the label or target.
- If the label is a number, such as yield in bags or a price in cedis, the task is regression.
- If the label is a category, such as repaid or defaulted, the task is classification.
There is also unsupervised learning, where there is no label and the model looks for structure, such as groups of similar customers. It appears as an optional topic in this section.