Fault diagnosis (FD) plays a crucial role in the industrial domain. Data preprocessing, particularly attribute reduction (AR), constitutes a pivotal aspect of FD methodologies in the present. Decision-theoretic rough set (DTRS) is a common method of AR. However, when employing DTRS for reducing attributes of continuous data, data discretization can inadvertently partition originally identical values into disparate intervals, thereby potentially impinging upon the ultimate reduction outcomes. Consequently, this study proposes an AR approach tailored for single-parameter decision-theoretic rough set (SPDTRS), leveraging fuzzy conditional probability to enhance both computational efficiency and accuracy. The proposed method entails sample classification predicated on similarities, followed by the computation of their fuzzy conditional probabilities to derive a matrix of loss functions oriented towards data. Subsequently, AR is executed utilizing SPDTRS, guided by the principle of global risk minimization. Based on the proposed AR method, a FD strategy is presented. Comparative experiments are conducted utilizing both the UCI data set and satellite power system FD scenarios. The results showcase the efficacy and dominance of this approach.