An Adaptive Probabilistic and Sparse Learning Paradigm for High Dimensional Fault Diagnosis in Industrial Cyber-Physical Systems

Abstract

The rapid growth of Industrial Internet of Things (IoT) technologies has enabled continuous monitoring 
of industrial processes through large volumes of sensor-generated data. Traditional monitoring systems 
primarily relied on manual inspection and rule-based approaches, which often struggled to provide real
time insights and efficiently handle complex industrial environments. To address these limitations, this 
study presents an intelligent industrial analytics framework based on Machine Learning (ML) 
techniques integrated within a web-based platform. The framework employs a unified Two 
Classification and Two Regression Tasks (2CA2RT) approach to predict maintenance flags and 
production status through classification models, while simultaneously estimating downtime and 
efficiency scores using regression models. Multiple algorithms, including Passive Aggressive (PA), 
Extreme Gradient Boosting (XGB), Adaptive Boosting (AB), Neural Architecture Search–Greedy Rule 
Forest (NAS-GRF), and the proposed Probabilistic Neural Network with Scope Rule Model (PNN
SRM), are implemented and evaluated using industrial sensor data. The proposed PNN-SRM model 
combines probabilistic learning with sparse rule optimization to improve predictive performance in 
high-dimensional industrial datasets. The framework also incorporates data preprocessing, Exploratory 
Data Analysis (EDA), model training, evaluation, visualization, and prediction modules to provide a 
complete end-to-end solution. Experimental results demonstrate that the proposed PNN-SRM model 
outperforms all baseline and hybrid models, achieving 100% accuracy (1.0000) for both maintenance 
flag and production status prediction tasks. Additionally, it attains superior regression performance with 
R² scores of 99.79% (0.9979) for downtime prediction and 99.96% (0.9996) for efficiency score 
prediction. These results highlight the effectiveness of the proposed framework in enhancing predictive 
maintenance, reducing operational downtime, and supporting intelligent industrial automation systems