Dynamic Query Algorithms for Human-Computer Interaction Based on Information Gain and the Multi-Layer Perceptron

Algorithms are presented for an "intelligent" human machine interface for efficient on-line decision making and pattern recognition. The algorithms structure the data input process by dynamically asking the user the next most "informative" question based on its current state of knowledge, to reach a conclusion as quickly as possible. Using the information gain principle in attribute selection, IQA and IQA1 dynamically generate a query process without the construction of the decision tree. A further development, IQA2, generalises IQA1 by including a multi-layer Perceptron (MLP) to mitigate the effect of noise and ambiguity and to establish incremental learning. The IQA algorithms perform well on noisy and incomplete data-sets. This is demonstrated by an example from an artificial domain and two form medical diagnosis.