Industrial cost is a comprehensive indicator to reflect industrial behaviors, whereas industrial data with annotation are rare because the annotation process is very expensive. Data increment transformation is a feasible solution to enrich annotated industrial data, but it bring a new challenge in system modeling because the size of transformational data is the quadratical relationship with that of collected data, and even turn into big data problem. Hence, a novel rule-based system proposed for handling big data problems, called micro-extended belief rule-based system (Micro-EBRBS), is introduced for industrial cost prediction. Firstly, the Micro-EBRBS is improved by 1) the use of activation factor to revise the calculation of individual matching degrees; 2) the use of parameter optimization to determine the optimal value of basic parameters. Afterwards, on the basis of data increment transformation, a novel industrial cost prediction model, called data increment- based Micro-EBRBS (DIME) model, is developed to accurately predict industrial costs. In case study, 13 state-own holding industries with historical data from 1999 to 2019 in China are used to illustrate the effectiveness of the DIME model. Comparative results show that the DIME model is more accurate than some existing models in industrial cost prediction.
|Journal||International Journal of Machine Learning and Cybernetics|
|Publication status||Accepted/In press - 25 Nov 2021|