Applying Predictive Analytics to Process Safety Leading Indicators

Abstract

Leading indicators can be defined as safety-related variables that proactively measure organizational characteristics with the intention of predicting, and subsequently avoiding, process safety incidents. Leading indicators become especially powerful when combined with advanced statistical methods, including predictive analytics. Predictive analytics is a broad field encompassing aspects of various disciplines, including machine learning, artificial intelligence, statistics, and data mining. This paper presents a case study in applying predictive analytics. Methods: The author developed a predictive derailment model for railroad application that could be modified and applied to process industries. Using regularly updated inspection data, the model was created using a logistic regression modified by Firth’s penalized likelihood method due to the low ratio of events to misses (i.e. track miles with derailments compared to track miles without derailments). The resulting model provides derailment probabilities for each mile of track over a six-month period. Additionally, the model identifies the variables that are significantly contributing to derailments, thereby showing the company which factors to address to prevent future incidents. Results: Model validation revealed that it demonstrated statistically significant predictive ability for 75% of derailments. Discussion: The same methodology could be used in the process industries to predict and prevent incidents, provided that organizations: 1. Identify leading indicators with predictive validity 2. Measure indicators at regular intervals 3. Create a predictive model based on measured indicators 4. Deploy the model whenever leading indicator measurements are taken to calculate predicted incident probabilities