Experimental and predictive analysis of pile raft foundation in clayey soils using machine learning and MATLAB tools

This study demonstrates the performance of piled raft foundations in a variety of clay soils collected from four states in southern India. Comprehensive physical, chemical, mineralogical and geotechnical characterisation of 12 different soil samples were performed. The testing on 12 soils reviles that soil has 54 to 70% clay content with predominantly montmorillonite mineral, which leads to high FSI values between 75 to 200%, and low dry densities in the range of 13.5 to 16.9 g/cc. Experimental loading tests were executed with different piled raft configurations such as no pile (R), 4 piles (R4), 6 piles (R6), and 9 piles (R9) raft. These studies extracted data for the calculation of the Load Sharing Ratio (LSR) and Settlement Reduction Ratio (SRR) in relation to the standard shallow foundation on homogeneous clayey soil. A 64% increase in LSR and 75% reduction in settlement values were observed as the number of piles increased from R4 to R9 sequentially. Machine learning (ML) models were developed to predict LSR and SRR based on soil parameters such as clay percentage, soil type, Free Swell Index (FSI), and Maximum Dry Density (MDD). The trained ML models successfully predict the LSR and SRR and for higher configurations with 12 (R12), 15 (R15), 18 (R18), and 21 (R21) piles with an accuracy of more than 90%. Finally, the LSR and SRR values obtained from ML models of experimental data for the test were compared via MATLAB. This verification process confirmed the high (> 95%) accuracy of the ML models in all scenarios. This study provides a broad direction on the designing of pile raft in a cost effective manner for a wide range of clayey soils.