Light pavement rehabilitations and low-cost treatments are extensively employed among transportation agencies on roads with relatively low traffic volumes to optimize available resources. One concern with this approach entails the difficulties of determining the optimal timing for treatment application. Making the best use of limited resources requires improvements in maintenance decision-making for selecting treatments considering all affecting factors and previous experience. This paper presents a machine learning approach in the decision-making process for determining the most appropriate pavement maintenance and rehabilitation alternatives for low-volume paved roads at the network level. Based on regional experts’ recommendations and engineering judgments in Colorado, a wide range of 884 cases of pavement-treatment patterns were generated. Then an artificial neural network (ANN) was trained with pattern-recognition algorithms. Two ANN prediction models were developed on the basis of pavement condition data, represented by six condition indices, and road lengths. The objective of training the models is to evaluate the variability of maintenance practices among five engineering regions within the Colorado Department of Transportation (CDOT). The outcome of this study describes the implementation gaps of pavement-preservation activities among CDOT regions resulting from limited maintenance funding. The regional maintenance selection can be processed by the developed ANN decision-making tool to recommend alternatives from regional recommendations as well as similar applications statewide to fit pavement management needs and expected performance.