Evaluation of Different Optimizers in Neural Networks with Imbalanced Dataset

Abstract

-With an increasing number of embedded sensor systems and data collection units set up in production plants, machines, cars, etc., there are new possibilities to store, analyse and monitor the data from such systems. These development makes it possible to detect anomalies and predict the failures that affect availability of these systems and impact maintenance plans. Typical industry scenario points towards have very less failures and data points related to same being captured in systems making it difficult to predict a rare event. This paper would be focusing towards evaluating the different optimizers and impact they have on accuracy while trying to predict a rare event target in a time series-based data. We would be evaluating different built-in optimizer classes in by tensor flow for training neural networks.