Membership Functions

The membership functions can be derived from the raw data by one of two methods:

• Equally spaced, where the intervals are decided based on points that equally separate the maximum and minimum seen values in the column
• Distribution based, where the membership sets are defined using the feature distribution. This is the default method.

The difference in the fuzzy set generated using the two options is visible in the following two images, the first being equally spaced, the second distribution based.

The platform uses type-2 fuzzy logic: each feature in the dataset has both a lower membership function and an upper membership function. At values where the lower function is strictly less than the upper function (rather than equal), the difference represents uncertainty in the membership value.

The info tab of a feature shows the boundary values for the core, lower membership function and upper membership function.

Hovering over the fuzzy set will show the boundary values in a popup.

The default number of fuzzy sets is three, which is derived from splitting the population into heptiles (seven partitions), however, this can be adjusted to have more or less fuzzy membership functions at the modelling stage.