Survival Analysis on Structured Data using Deep Reinforcement Learning

TL;DR

This paper introduces a Deep Reinforcement Learning approach using Double Deep Q Network to improve device failure prediction in predictive maintenance, especially under varying input data conditions, outperforming existing models.

Contribution

The paper proposes a novel DDQN-based method for survival analysis that handles input data variation better than traditional deep learning models.

Findings

The DDQN model predicts device failure more accurately than existing models.

The proposed method performs well with limited training data.

It demonstrates robustness to input data variation.

Abstract

Survival analysis is playing a major role in manufacturing sector by analyzing occurrence of any unwanted event based on the input data. Predictive maintenance, which is a part of survival analysis, helps to find any device failure based on the current incoming data from different sensor or any equipment. Deep learning techniques were used to automate the predictive maintenance problem to some extent, but they are not very helpful in predicting the device failure for the input data which the algorithm had not learned. Since neural network predicts the output based on previous learned input features, it cannot perform well when there is more variation in input features. Performance of the model is degraded with the occurrence of changes in input data and finally the algorithm fails in predicting the device failure. This problem can be solved by our proposed method where the algorithm can predict the device failure more precisely than the existing deep learning algorithms. The proposed solution involves implementation of Deep Reinforcement Learning algorithm called Double Deep Q Network (DDQN) for classifying the device failure based on the input features. The algorithm is capable of learning different variation of the input feature and is robust in predicting whether the device will fail or not based on the input data. The proposed DDQN model is trained with limited or lesser amount of input data. The trained model predicted larger amount of test data efficiently and performed well compared to other deep learning and machine learning models.