Using ML Methods for obtaining the prediction of maximum dry density

In geotechnical engineering, maximum dry density (MDD) is a fundamental concept, referring to the maximum mass per unit volume under specified conditions. MDD is pivotal for ensuring the stability of earthwork, including foundations and dams. It depends on factors like moisture, compaction force, grain size, and soil type. Accurate MDD prediction helps engineers ensure durability and reliability in civil engineering projects over time. This article, an original prediction method of MDD by means of the K-nearest neighbors (KNN) algorithm has been presented. Using the KNN method, accurate models can be developed that connect the MDD of treated soil to diverse natural soil attributes, including linear shrinkage, plasticity, particle size dispersion, and the type and content of stabilization additives. A large database comprising 187 samples representing various categories of soils from previously conducted stability tests has been utilized in this study to develop forecasting models and validate them. Besides, application of meta-heuristic strategies, specifically the Jellyfish Search Optimizer (JFO) and Black Widow Optimizing Algorithm (BWOA), further enhances the accuracy level of the KNN method in this work. Therefore, two hybrid models, KNJF and KNBW, are defined. Then, the data for the training, validation, and tests show excellent R2 values of 0.9895, 0.9681 and 0.9659, respectively, by the KNJF model. In addition, during the training phase KNJF has the best RMSE of 24.236. Everything taken into account, the KNJF model generalizes and predicts better than the models KNN and KNBW developed specifically for this investigation.