Introduction

ARFF was developed by the Machine Learning Project at the University of Waikato for use with the Weka machine learning software. It is a human-readable ASCII text format that describes instances sharing a fixed set of attributes.

An ARFF file has two sections: the Header (relation name and attribute declarations) and the Data section (one instance per line, comma-separated).

Structure

The @relation line names the dataset. Each @attribute line declares one attribute by name and type. The class attribute is conventionally last.

Attribute Types

For the full specification see the official Weka ARFF documentation.

Missing Values

Missing values are represented by a question mark ? in the data section. The Preprocess tab reports missing-value counts per attribute. During classification and clustering, missing values receive a maximum-distance penalty.