Estimation vs Prediction Vs Forecast

Question

I already read the q/a's from here and here. These answers are quite a bit complicated for me. I need something easy to understand and explain to any interviewer. I read lots of articles like;

What is the difference between estimation, extrapolation, prediction, and forecasting?

Prediction vs Forecasting

What is an Estimate

and lots more...

What I got from these articles in a nutshell is:

Estimation

What: It approximates the result based upon historical understanding and experience.

When do we need to estimate: If we don’t have data to support an exact result, then in such case we do a rough estimation.

where do we use: Financial transaction, Estimating the cost of the item which has a price tag of “$10 + Vat”, Estimating the number of people who could attend the party, Estimate current time without looking to the watch and so on.

Prediction

What: It gives output close to the actual answer based upon past data but ignores certain conditions.

When do we need to predict: To have a sense of control

Where do we use: Fraud detection, Content recommendation, Health diagnosis, House price prediction, and so on

Forecast

What: It gives output closest to the actual answer that may possibly happen in the future. It is also a type of prediction. It predicts what will happen in the future by taking into consideration events in the past and present both

When do we need to forecast: To predict the possible future with high accuracy.

Where do we use: Business Planning, Budgeting, Investing money in any business wisely, Weather forecasting, and so on

I tried to explain them very easily but still, I feel like most of the explanation is vague. Could anyone help me to make my explanation easy, clear, and correct?

Answer 1

As you learned from the two threads that you mentioned, in statistics estimation is about learning something about the data, while prediction is about predicting. It is not about "approximating" because every statistical method uses a mathematical model that approximates reality. Forecasting is a special case of making predictions, where we make predictions about the future.

Estimation

What: It approximates the result based upon historical understanding and experience.

When do we need to estimate: If we don’t have data to support an exact result, then in such case we do a rough estimation.

This is incorrect. When estimating, you are learning something that is not directly observable. You are not estimating the data, but inferring the properties of the data (see below).

where do we use: Financial transaction, Estimating the cost of the item which has a price tag of “$10 + Vat”,

Using your example: you treat the cost of an item as a random variable, that is characterized by a probability distribution, you want to learn characteristics of the distribution so that you can infer things like "average cost" or "the 95% quantile of the cost".

When you want to make a "guess", in statistics it's called making predictions. So colloquial "rough estimate" in statistics would rather be a "rough prediction".

Estimating the number of people who could attend the party,

This would be a forecast.

Estimate current time without looking to the watch and so on.

This would be a prediction.

Prediction

What: It gives output close to the actual answer based upon past data but ignores certain conditions.

Every mathematical model "ignores certain conditions" so this would apply to each of them. The "past data" is used to learn the predictive algorithm, but it makes the predictions using the current data. The usual scenario is that you have historical labels $y$ and features $X$, and you learn some function $y = f(X)$, so that yu can guess the unknown $y_\text{new}$'s if you have the $X_\text{new}$'s available.

Forecast

What: It gives output closest to the actual answer that may possibly happen in the future. It is also a type of prediction. [...]

Yes, it's like predictions, but with the difference that since it's about the future, you don't know the $X_\text{new}$'s because no data from the future can be observed. So when making regular predictions, you use partially observed data to guess something that is missing, while when forecasting you are using your historical knowledge to make guess on what could the future be. For example, with an RTG of someone's lungs, you can predict if they have lung cancer or you can forecast the chance of them getting lung cancer in the future.

Answer 2

They do have similar meanings. Actually they are used interchangeably sometimes.