 |
International Society of Science and Applied Technologies |
| Machine Learning-Based Spare Parts Demand Prediction under General Repair Turnaround Time Distributions | ||||
| Author |
Hongzhou Wang
|
|||
| Co-Author(s) |
Brian Hart
|
|||
| Abstract | Wang and Hart [1] introduced an Erlang C -based spare parts prediction model that addresses intermittent usage and prevents lost demand. It allows infinite spare parts requests, making it theoretically optimal. However, in practice, the repair turnaround time (RTAT) may not follow an Exponential distribution but a Lognormal distribution [5]. This paper will explore spares prediction when RTAT does not follow the exponential distribution which is assumed in the Erlang C queueing. Then this paper uses machine learning to refine the parameter of the spares prediction models, covering data preprocessing, feature engineering, train and test data preparation, variable importance, class balance, confounding, and the performance of various machine learning algorithms, including gradient boosting machine, distributed random forest, deep learning, and stacked ensemble. This paper has developed a machine learning algorithm with excellent performance. The length of this paper limits the discussions to prediction of repair outcomes only.
|
|||
| Keywords | Spare Parts; Machine Learning; Queueing theory; M/M/c Queue, M/G/c Queue; Erlang C, Deep Learning | |||
| Article #: RQD2025-34 | ||||
Proceedings of 30th ISSAT International Conference on Reliability & Quality in Design |