ARIMAX model for forecasting maintenance work (AMFM): A multi-stage seasonal ARIMAX model for work order time series forecasting

E-commerce business depends on smooth day to day functioning of its warehouses/facilities. The functioning of these facilities depend on the health of material handling equipment. To keep these equipment healthy, these facilities employ the help of maintenance engineers who perform predictive/breakdown maintenance work. To ensure an effective maintenance operation necessitates efficient planning of maintenance work. For efficient planning of future maintenance work, one needs to have good estimates of the future maintenance work. We created time series of maintenance work (breakdown and miscellaneous) in terms of demand hours for every day/week. Next, we built several models and evaluated these models on the basis of forecasting accuracy metrics viz. Mean Absolute percent error (MAPE) and Root mean squared error (RMSE) to determine which modelling technique is most suitable. Seasonal ARIMA with exogenous variable (SARIMAX) was found to be the most suitable approach with additional hyperparameters like training dataset length and training data window start/end. This paper discusses the details of this SARIMAX approach and the procedure used to identify the best facility specific SARIMAX model. The proposed solution provides forecasts using SARIMAX framework with an out of sample MAPE less than 30 percent and RMSPE less than 20.