The starting point for any machine learning exercise is the ultimate use case it aims to address. Sometimes, this goal will be statistical inference in order to identify an association between variables or even a causal relationship. Most frequently, however, the goal will be the direct prediction of an outcome to yield a trading signal.

Both inference and prediction use metrics to evaluate how well a model achieves its objective. We will focus on common objective functions and the corresponding error metrics for predictive models that can be distinguished by the variable type of the output: continuous output variables imply a regression problem, categorical variables imply classification, and the special case of ordered categorical variables implies ranking problems.

The problem may be the efficient combination of several alpha factors and could be framed as a regression problem that aims to predict returns, a binary classification problem that aims to predict the direction of future price movements, or a multiclass problem that aims to assign stocks to various performance classes. In the following section, we will introduce these objectives and look at how to measure and interpret related error metrics.