Road Construction Assessment Model (Rc-Am) to Prevent Contract Overbilling in Angola

Abstract

Despite the $U.S. 2.1 billion spent annually on road construction, quality of the roads in Angola is still poor in part due to existing issue of contract overbilling. To reduce road-project contract overbilling, this Praxis investigated the use of satellite imagery, which has not been extensively explored, to identify road layers and measure road lengths as triggers for road inspections. These triggers were predicted on the basis of spectral, spatial, and textural features extracted from randomly selected groups of pixels obtained from high-resolution (0.5 m) optical satellite imagery. To establish a model for these triggers, samples of pixels representing road segments were assigned to either training or validation groups. Machine-learning methods were then applied to classify road layers, using decision criteria for accuracy. Once the road layers were identified and the road lengths measured, the model could be used to detect overbilling and trigger inspections in different road construction projects.

The results showed good accuracy in the classification of road layers, with the kappa coefficients greater than 0.81 (almost perfect) and an overall accuracy exceeding the recommended threshold of 0.85. In addition, on the basis of the measured road lengths and road-layer detection, the completeness – a ratio between the measured and the reported lengths of road layers – was above the acceptable threshold of 0.95 for model validation, and below the threshold for some projects in model implementation, indicating potential contract overbilling that required onsite inspection. The greatest overbilling cost that our model predicted was $113,013.56 (corresponding to 20% overbilling), which requires the immediate attention of the inspector to validate the exact amount. Taken together, this is a promising model that has the potential to solve the problem of road-project contract overbilling in Angola.