Conditional Status Detection for Factory Management System with Optimized Predictive Modeling

Abstract

The idea is to build a wireless factory management system which can perform surveillance of the premises and report abnormal conditions by sensing parametric deviations in temperature, pressure, humidity, wind speed, and visibility using microcontrollers and sensor-based remote terminal unit (RTU). In this paper, two constraint functions were constructed which evaluate the conditional status of the site in terms of abnormality and help in the classification of the variables. Various machine learning algorithms like logistic regression, neural network, naive Bayes, random forest and gradient boosting with hyperparameter tuning are used, which predict the conditional outcome with a 95% accuracy. Also, the constraint functions are compared by various statistically and graphically parameters such as KS statistic, ROC curve and AUC, gain, lift and actual versus predicted charts. The results indicate an optimization of 2% in the accuracy of the predictive model by using the weighted percentage method.