Machine learning and optimization-based modeling for asset management: a case study

Purpose - This research is inspired by a real case study from a pump rental business company across the US. The company was looking to increase the utilization of its rental assets while, at the same time, keeping the cost of fleet mobilization as efficient as possible. However, decisions for asset movement between branches were largely arranged between individual branch managers on an as-needed basis. Design/methodology/approach - The authors propose an improvement for the company's asset management practice by modeling an integrated decision tool which involves evaluation of several machine learning algorithms for demand prediction and mathematical optimization for a centrally-planned asset allocation. Findings - The authors found that a feed-forward neural network (FNN) model with single hidden layer is the best performing predictor for the company's intermittent product demand and the optimization model is proven to prescribe the most efficient asset allocation given the demand prediction from FNN model. Practical implications - The implementation of this new tool will close the gap between the company's current and desired future level of operational performance and consequently increase its competitiveness Originality/value - The results show a superior prediction performance by a feed-forward neural network model and an efficient allocation decision prescribed by the optimization model.