Segregation (hot-mix asphalt segregation) is one of the main problems affecting asphalt pavement performance. The early detection is important, but the tests are quite expensive and time-consuming. The visual examination is the cheapest method but too varied in judgement and can rise further problems. In this experiment, we developed machine learning linear and quadratic discriminant analyses to detect/classify segregated and non-segregated pavement asphalt. Six variables were employed: SD only, IR only, MAD only, IR-mean, MAD-mean, IR-mean, MAD-SD-mean and IR-SD-mean. The results showed that the complexities of information affect machine learning performance. IR-SD-mean and MAD-SD-mean parameters gave best accuracy performance for training data at 99.2% (LDA)/98.5% (QDA) and testing data at 98.33% (LDA)/95% (QDA) respectively. In general, QDA gave more accuracy performance in comparison to LDA although our data dimension is small.
Keywords: Asphalt pavement segregation; Machine Learning; segregation detection; Linear discriminant analysis; quadratic discriminant analyses