Understanding the material properties of a pavement structure is crucial for evaluating the pavement’s performance and assessing its damage level. Generally, the backcalculation process is extensively used to analyze the Falling Weight Deflectometer (FWD)-data for estimating the layer-moduli of a pavement structure. It is mainly an iterative process that starts with a set of seed (initial) variables, calculates the theoretical pavement surface deflections, and compares them to the measured deflections. Yet, this process is most likely unstable and is prone to numerous errors including the selection of relevant seed variables. The selected seed-variables hold significant consequences on the-final backcalculated-results. This research project aims-to-develop models through classification analysis to predict the seed variables. This involves (1) calculating theoretical surface deflections through a finite element model that simulates different pavement structures and properties, (2) calculating FWD parameters and indices for each structure and (3) using those parameters to build Random Forest models that predict the seed variables with low OOB-error and high accuracy. The dynamic approach is adopted to perform the analysis on 3-layered rigid and flexible pavements. The AC layer is modeled as an LVE/material while the PCC and the unbound layers are modeled as linear/elastic materials with damping. The OOB-Estimate of error rate and the overall accuracy values obtained dictate that the predictor variables selected to build the RF models are efficiently trained and generate accurate predictions for all seed variables except for the Rayleigh Damping Parameter of the PCC layer “𝛼𝑅𝑃𝐶𝐶”. The developed models can be considered as an effective guidance for pavement engineers to select the seed variables that are closer to the actual values to initiate the backcalculation process.