Semi-Supervised Learning for Sensor-Based Flash Point Prediction in Oil Industry

The limited availability of labeled data is a significant problem for improving the performance of machine learning models. To address this, semi-supervised learning methods that leverage both labeled and unlabeled data are increasingly adopted. These methods utilize the inherent structure of unlabeled data to improve learning accuracy and are particularly beneficial in situations where labeled data are scarce or expensive to obtain. This study investigates the efficacy o f semi-supervised learning techniques in solving a critical problem, specifically the prediction of flash points in the oil industry, where limited labeled data are available. Accurate prediction of flash point, crucial for the safe transport and processing of oil products, is presented as an efficient alternative to traditional methods in terms of time and cost. This research was conducted on a new and real dataset containing flash point va lues derived from sensors' data on pressure, temperature, and flow indicators, along with laboratory measurements. Various preprocessing techniques, such as Winsorization and Min-Max Scaling, were applied to the dataset. The semisupervised learning based model has provided acceptably accurate predictions even in situations with limited labeled data. The application of this approach allows for the improvement of operational efficiency and safety i n various fields where limited labeled data a represent, including the oil and chemical industries, health, finance, and environmental monitoring. The findings of the research demonstrate how semi-supervised learning techniques can contribute to the optimization of processes by enhancing overall prediction accuracy.