Track: Data Analytics

Abstract

Logical Analysis of Data (LAD) is a classification algorithm known for its high accuracy and interpretable results, but it has relatively long computational time, which makes it unsuitable to treat big data. In this paper, we propose a platform for improving the computational efficiency of LAD especially in large scale and big data applications. We develop an approach based on a merging operation of two different LAD models. The proposed approach is ideal to develop incremental learning platform based on LAD. In addition, it enables LAD to be performed in parallel computing environments. The suggested operation determines the intersection between each pair of patterns from the two models. Each intersection is a new pattern in the new merged model and its characteristics are inherited from its parent patterns. Numerical experiments are conducted by using an industrial data set. The experiments demonstrate the trade-off between accuracy and computational efficiency of the proposed approach and the classical implementation of LAD with a sole run on the whole data. The proposed approach resulted in a maximum accuracy reduction of 5% and computational time reduction of 87% from the accuracy and computational time of the classical implementation of LAD, respectively.