Why this matters
Many models used by Ghanaian companies output a probability rather than a plain answer: how likely a loan is to be repaid, how likely a mobile money transfer is fraud, how likely a customer is to stop using a service. Someone must turn that number into a decision, and someone must explain what it means. Data science adverts name probability for this reason.
Probability in one line
A probability is a number from 0 to 1 that says how likely something is. 0 means it never happens, 1 means it always happens, and 0.25 means it happens about one time in four over many similar cases. A single prediction of 0.9 can still turn out wrong. It only promises that, across many cases given 0.9, about nine in ten should happen, and only if the model is well calibrated, meaning its numbers have been checked against real outcomes.
Conditional probability
Most useful questions are conditional: how likely is A, given that we know B? We write it as P(A | B), read "the probability of A given B". The chance a maize farm gets a good harvest is one number. The chance it gets a good harvest given that the rains came early is a different, more useful number. A classification model is really estimating a conditional probability: the chance of a label given the inputs.
The base rate trap
Here is an invented example that catches many beginners. Out of 1,000 mobile money transfers, 10 are fraud. A model flags 9 of those 10 fraud cases, which sounds great. But it also wrongly flags 50 of the 990 honest transfers.