Strength characteristics modeling of lateritic soils using adaptive neural networks

Laterite, as a soil group rather than a particular material, is more commonly found in the leached soils of humid tropics. It is difficult to assess the strength parameters of lateritic soils from the existing literature due to variability in test methods, degree of compaction, and actual soil characteristics. The California Bearing Ratio (CBR) has been suggested as a more realistic basis for evaluating the quality and strength of lateritic soils and gravels for their use in pavement construction. This paper examines the use of neural networks, especially, generalized adaptive neural networks, for modeling the field strength characteristics based on CBR, and compares the results to a more traditional back-propagation neural network approach. It is found that both methods are fairly applicable in strength characteristics modeling of lateritic soils. However, the generalized adaptive neural network seems to be more successful in its predictive capability, and demonstrates that few input variables are needed to determine the strength characteristics of lateritic soils. The input data to the adaptive neural network include the index properties, field density and moisture content. (C) 1997 Elsevier Science Ltd.