A hybrid XAI-FCI approach for job cycle time range estimation: wafer fabrication as an example

Abstract Fuzzified deep learning (DL) models have been widely used to estimate the job cycle time range to contain the actual value. However, this is a challenging task for DL models with complex structures and operations. To overcome this challenge, a hybrid explainable artificial intelligence (XAI)-fuzzy collaborative intelligence (FCI) approach is proposed in this study for job cycle time range estimation. In the proposed methodology, experts collaborate to predict the job cycle time using DL models. Each DL model is then explained by a random forest (RF), and the RFs of all DL models collaborate to fuzzify the DL output to estimate the job cycle time range. In this way, the DL model is fuzzified precisely in a post hoc way regardless of its complex structure and operations. Subsequently, another FCI mechanism is used to aggregate the job cycle time ranges estimated by all experts. The proposed methodology is novel as it is the first attempt to combine XAI and FCI. In addition, the first FCI layer collaborates multiple decision trees, while the second FCI layer collaborates between experts. Furthermore, the two FCI layers collaborates using most possible range (MPR) and partial-consensus fuzzy intersection (PCFI), respectively. The hybrid XAI-FCI approach has been applied to a real case. According to the experimental results, the hybrid XAI-FCI improved the precision of estimating the job cycle time range by up to 41%.