Overview
In a feature based forecasting system, the analyst builds a model that tries to forecast the rating that a customer would give to a target product based on the set of features of the customer and product that are available to the analyst (such as age, gender, education, income etc.....)
Building this type of system is fairly straightforward, given a set of customers who have already provided ratings for the product, a model is built, which could include any of the following.
- Binary Model
- a binary model is a model with eseential two ratings or outcomes. The outcomes could be labeld
like and dislike, or purchase and not purchase.
A Logistic Regression is a linear model that returns a number between {% 0 %} and {% 1 %}. This would indicate a probability that the customer will like or purchase the product. - Multiple Ratings
- when the product has a rating with more than 2 labels, the analyst can build a model to accommodate the larger
set of labels using the following:
- Linear Regression
- builds model of the form
{% rating = \alpha + \beta_1 Feature_1 + ... + \beta_n Feature_n %}
- A Neural Network model can also be constructed with the feature vector as an input, and the rating as the output. Because numeric ratings have a natural ordering relation, the output can be structured as a single output, but the activation function on the output must not be bounded within the rating spectrum.
- Linear Regression
- builds model of the form