Semi-supervised Learning for Mixed-Type Data via Formal Concept Analysis

• We propose a semi-supervised learning (SSL) method, called SELF (SEmi-supervised Learning via FCA), using Formal Concept Analysis (FCA) – It can handle mixed-type data containing both discrete and continuous variables ∘ Numerical data are discretized by binary encoding / Summary • We propose a semi-supervised learning (SSL) method, called SELF (SEmi-supervised Learning via FCA), using Formal Concept Analysis (FCA) – It can handle mixed-type data containing both discrete and continuous variables ∘ Numerical data are discretized by binary encoding • Main contributions . The first direct SSL method for mixed-type data – FCA is shown to be a powerful tool for machine learning and knowledge discovery . Can handle incomplete datasets – missing values and missing labels . Achieve good accuracy of classification experimentally /