In this section, we will look at what terms are used in forecasting and what they mean. Here is a quick scenario where we will be predicting customer lifetime value with the machine learning model. Let's predict the lifetime value of an e-commerce customer using regression. Our goal is to better target high-value customers to our e-commerce site throughout the customer lifecycle with special promotions and incentives. After exploring the data, we can provide a number of useful fields to the model, like the number of different days a visitor has been to our website. How many lifetime page views, how many total visits, what is the average time on site, the total revenue brought in, and the count of e-commerce transactions on our site. Now, all that is shown here is basic analytics to get a sense of the data, but you could feed this historical lifetime value data and use that to predict and power which customers are high-value customers to help you focus and target them for promotions and incentives. But before we get too deep building models in BigQuery, we first need to define our data terms in the language that data scientists and other machine learning professionals use. Taking the e-commerce example we had in the previous lesson, a record or a row is called an instance or an observation. In the screenshot you see here, we have eight instances. A label is the correct answer and will be what you are looking to train the model on with your existing data and predicting with your model on future data. Here the label is Lifetime Revenue, which is a number we will be trying to predict. Labels could also be things like binary values like high value customer or not as shown here. Knowing what you're trying to predict a class, a number, etc will greatly influence the type of model you will use later. Those columns are called features. We have a whole module dedicated to creating machine learning data sets in BigQuery which touches on the critical topic of feature engineering, which is exploring, cleaning, and pre-processing your data before you input it into your machine learning model. This is often the hardest part of any ML project and why it's great you already enjoy working with data as analysts. Now, say some new data comes in that you don't have a label for, we now have a data set of labeled examples and a data set with some unknowns. Well, this is the fun part, we can draw inference or predict those values with a model. Again, a machine learning model will build a recipe for determining those output values in this case classifying whether that customer is high value or not based on your labeled training data, which is the blue box shown here. Next up, you'll learn how to code these models yourself with just SQL. Your model will learn the way to give each feature as you can see on the screen. Some features such as distinct days visited have more weight than average session quality. After the model is trained you can see the relative importance of each field. A label is the correct answer and will be what you are looking to train the model on with your existing data and predict with your model and future data. Here the label is Lifetime Revenue, which is a number we will be trying to predict. Now, let's say you have to predict future values of lifetime value for a customer. We now will use past values of LTV for that customer to predict future value. Now say some new data comes in that you don't have a label for, we now have a data set of labeled examples and a data set with unknowns.