Model Selection Process in Time Series Analysis of Production System with Random Output

Abstract

In time series investigation of characteristics of production system, different competing models are generally obtained particularly in production settings with stochastic output attributable to bottleneck problems. Consequently, selecting the best model that describes a production system becomes challenging and critical because some models that fit observed data most accurately may not predict future values correctly on account to model complexities. This research desires to demonstrate the procedure for model selection in production system with random output via the use of Adjusted Coefficient of Determination (), Akaike and Schwarz criteria tools. Production output measurements obtained serve as input data to Autocorrelation Function and Partial Autocorrelation Function to obtain the order of Autoregressive, Autoregressive Moving Average and Autoregressive Integrated Moving Average models. The model parameters were estimated and used for predictions and compared with original and transformed data to obtain Sum of Squared Error (SSE). Afterward, the models were subjected to adequacy evaluation and subsequently tested with Akaike and Schwarz criteria. Among the competing models, ARIMA (3, 1, 1) model explain 66% variance of the dataset and wielded the lowest Akaike and Schwarz values of 534.41m and 534.34m respectively and thus selected as the model that represents the production system under investigation. The approach establishes that Adjusted Coefficient of Determination in conjunction with Akaike and Schwarz criteria are adequate tools for model selection in time series investigation particularly in stochastic situation