V3I11P3

Benchmarking the Extra-Trees and Hybrid Physics–Informed Models against Classical CPT-Based Methods for Shallow Foundation Settlements on Lateritic Soils

Njah, C. S^1*, Abam, T. K. S², Cliff-Teme, S³, Jaja, G. W. T¹

Abstract

Terzaghi’s consolidation theory, Janbu’s tangent-modulus approach, and the CPT-based correlations proposed by Meyerhof and Schmertmann remain widely used for estimating foundation settlement. Their original calibrations, however, were based largely on temperate soils or clean granular deposits, so their performance in tropical lateritic soils needs to be examined rather than assumed. This study evaluates six approaches: a tuned Extra Trees ensemble (the XT-Settlement Model), the closed-form Hybrid Physics–Informed Settlement Model (HPISM), and four classical methods. The comparison uses three CPT parameter sets representing typical, favourable, and adverse lateritic ground conditions. On the independent test set of 2,681 records, the tuned ensemble achieved R² = 0.9950; the other data-driven formulations recorded R² values between 0.44 and 0.99. The worked examples show increasing disagreement among the methods as the conditions depart from the database mean. Meyerhof’s correlation, developed for clean sands, returned settlements of only 2–7 mm at the low cone resistances considered, while the data-driven models predicted 38–309 mm. Schmertmann’s method gave the largest estimate under heavy loading (359 mm). Terzaghi’s and Janbu’s estimates were of a more comparable order, but both required parameters—such as void ratio and a constrained-modulus coefficient—not available from CPT measurements alone. These results indicate that methods calibrated on temperate granular soils should not be transferred to lateritic ground without independent validation. They also provide a documented basis for selecting a method for preliminary assessment and design.

Keywords:

Classical settlement methods; Comparative validation; Cone penetration test; Lateritic soils; Shallow foundation settlement