Monitoring operational efficiency of construction equipment, which indicates how efficiently construction equipment is utilized, provides key information in reducing air pollutant emissions from equipment use as well as improving the productivity of construction operations. In this paper, we report our efforts to measure the operational efficiency of construction equipment, using low-cost accelerometers. The measurement of the operational efficiency of construction equipment is formulated as a problem that classifies second-by-second equipment activity into working, idling, and engine-off modes. We extract various features from the raw accelerometer data and classify them into three different equipment activities (working, idling, and engine-off), using supervised learning algorithms such as Logical Regression, decision trees, k-Nearest Neighbor, and Naïve Bayes. The result from the real-world experiment indicates that the use of supervised learning algorithms provides over 93% of recognition accuracies, and this level of accuracies causes less than 2% error in the measurement of equipment operating efficiency.