Vol.:(0123456789) 1 3 International Journal of Geosynthetics and Ground Engineering (2021) 7:44 https://doi.org/10.1007/s40891-021-00282-x ORIGINAL PAPER Prediction of Ultimate Bearing Capacity of Aggregate Pier Reinforced Clay Using Machine Learning Sharad Dadhich 1  · Jitendra Kumar Sharma 1  · Madhav Madhira 2,3 Received: 4 January 2021 / Accepted: 28 April 2021 / Published online: 30 May 2021 © The Author(s), under exclusive licence to Springer Nature Switzerland AG 2021 Abstract Aggregate piers are extensively in use for increasing bearing pressure and diminish settlement under the footing. The ulti- mate bearing capacity of aggregate pier reinforced clay is majorly afected by soil strength (c u ), area replacement ratio (a r ) of piles, geometry, and slenderness ratio (λ) of piles. Various prediction models have been proposed to predict the ultimate bearing capacity of aggregate piers. However, existing models have shown a broad range of bias, variation, errors, and as such they are unsuitable for practical design. In this study, machine learning algorithms (linear and non-linear regression) and Artifcial neural networks (ANNs) were performed using feld loading test results to predict the ultimate bearing capacity of ground reinforced by aggregate piers. Sensitivity analysis was conducted to identify the infuence of input variables. To fulfl this objective, 37 test results were used for training and testing of diferent models and compared with each other based on statistical parameters (mean absolute error, root mean squared error, and r 2 -score). Random Forest Regression model came out to be the best suitable for prediction of ultimate bearing capacity with minimum mean absolute error (MAE = 38.93 kPa) and r 2 -score equal to 0.98. Keywords Aggregate pier · Ultimate bearing capacity · Machine learning · Artifcial neural networks · Sensitivity analysis Introduction Soft soils, such as clays and silts, have low strength and high compressibility. Thus, soft soils are vulnerable to construc- tion activities at moderate loads and need ground improve- ment before construction. Various ground improvement techniques such as soil-stabilization, pre-loading with PVDs, and stone columns/granular piles have been used extensively in small to medium projects. Recently, aggregate piers are being used extensively to increase bearing capacity and reduce settlement and lateral displacements under founda- tion. Aggregate piers act as vertical drains and accelerate the consolidation of surrounding soft soil. The prediction of the altered or modifed ground’s ultimate bearing capacity is an essential task for a proper design [1]. Since the early 1970s, many researchers had aimed to develop a methodol- ogy based on elastic and plastic theories [2], cavity expan- sion theory [3, 4], numerical [5, 6], and empirical methods [7, 8]. Predictive models were developed continuously and modifed. Laboratory and feld tests on the aggregate piers had been performed to investigate the failure mechanisms, and their ultimate bearing capacity evaluated [9–12]. Ambily and Gandhi [13] presented a detailed experimental study carried out on the behaviour of single and group of stone columns, and results from the study were compared with the fnite element analysis model. Hanna et al. [14] presented a numerical model to simulate single and group of stone- column performance in soft soils. Mohanty and Samantha [15] studied the behaviour of stone columns in the laboratory and proposed a numerical study. Algin and Gumus [16] pre- sented 3D numerical modelling considering the installation efect. Etezad et al. [17] developed an analytical model to predict soft soil bearing capacity reinforced with stone-col- umn and validated it via numerical and experimental result. In the early nineties, artifcial intelligence (AI), because of its capability and versatility, came in to picture to solve several critical civil engineering problems. Artifcial neural network (ANN) is a machine learning algorithm that simu- lates the working of personage brain by automation of data mining and data acquisition. Many researchers extensively used ANN to obtain solutions for many complex problems. Literature studies revealed that ANN had been used for Extended author information available on the last page of the article