Vol.:(0123456789) 1 3 Innovative Infrastructure Solutions (2023) 8:2 https://doi.org/10.1007/s41062-022-00966-x STATE-OF-THE-ART PAPER Modelling soil compaction parameters using a hybrid soft computing technique of LSSVM and symbiotic organisms search Lal Babu Tiwari 1  · Avijit Burman 1  · Pijush Samui 1 Received: 16 June 2022 / Accepted: 14 October 2022 / Published online: 2 November 2022 © Springer Nature Switzerland AG 2022 Abstract The soil compaction parameters, i.e., optimum water content (OWC) and maximum dry density (MDD) are essential param- eters used in civil engineering projects for monitoring the compaction of soils. The current practice of using laboratory testing to determine the OWC and MDD is time-consuming and costly. Thus, this research suggests a hybrid machine-learning solu- tion to replace traditional soil testing for determining OWC and MDD. The novel method combines the least square support vector machine (LSSVM) and symbiotic organisms search (SOS) algorithm. These two computational intelligence algorithms work together to create an OWC and soil MDD prediction model, LSSVM–SOS. For this purpose, a large database of 13 diferent soils featuring 6 infuencing factors was used. Overall, experimental results show that the proposed LSSVM–SOS has attained the most accurate prediction of the OWC of soils (RMSE = 0.0288, MAE = 0.0199, and R 2 = 0.9656) and MDD (RMSE = 0.0305, MAE = 0.0206, and R 2 = 0.9641). These results of the proposed model are signifcantly better than those obtained from other hybrid LSSVMs constructed with particle swarm optimization, grey wolf optimizer, and slime mould optimization algorithms. According to the fndings, the newly created LSSVM–SOS can aid geotechnical engineers in the design phase of civil engineering projects. Keywords Soil compaction · Support vector machine · Artifcial intelligence · Particle swarm optimization · Swarm intelligence Introduction Soil compaction is the process of pressing soil particles closer together by minimizing air voids while keeping the water content between the soil particles constant. The mechanical properties of soils can be improved in a variety of ways through compaction. Proctor [1] suggested compact- ing the soil at the desired compaction energy with varying water contents. The compaction curve can thus be used to determine the optimum water content (OWC) and maximum dry density (MDD). These two compaction parameters are frequently utilized in geotechnical practices to maintain the long-term performance of diferent geotechnical structures, such as highway embankments [2, 3], railway embankments [4, 5], airport runways [6–8], and so on [9]. For the construc- tion and maintenance of geotechnical structures, it is conse- quently essential to comprehend and predict the compaction characteristics of various soils [9, 10]. The OWC and MDD can be determined using laboratory experiments and analytical methods [10]. In the laboratory, at least 4–5 tests must be performed to accurately defne the compaction curve. Thus, the laboratory test is tedious and time-consuming [11]. In addition, veteran geotechnical experts and highly qualifed personnel are required to con- duct the test and attain reliable results. Hence, it is essential to develop intelligent data-driven algorithms for determin- ing the OWC and MDD based on available experimental records [9, 10]. In the past, several prediction models were proposed to determine OWC and MDD of soils. The major- ity of these models were developed using regression analy- ses and limited data from specifc soils. Wang and Yin [10] stated that these models produced a wide range of prediction accuracies, with coefcients of determination (R 2 ) scattered * Pijush Samui pijush@nitp.ac.in Lal Babu Tiwari lalbabut.ce@nitp.ac.in Avijit Burman avijit@nitp.ac.in 1 Department of Civil Engineering, National Institute of Technology Patna, Patna, Bihar, India