
AI Blueprints
By :

Identifying trends and anomalies involve similar techniques. In both cases, we must fit a model to the data. This model describes what is "normal" about the data. In order to discover trends, we'll fit a trend model for the data. A trend model fits a linear, quadratic, exponential, or another kind of trend to the data. If the data does not actually represent such a trend, the model will fit poorly. Thus, we must also ask how well a chosen model fits the data, and if it does not match the data sufficiently well, we should try another model. It's important to make this point because, for example, a linear trend model can be applied to any dataset, even those without linear trends (for example, a boom-and-bust cycle like bitcoin - USD prices between mid-2017 and mid-2018). The model will fit very poorly, yet we could still use it to predict future events – we will just likely be wrong about those future events.
In order to recognize anomalies, we take a model of what is...