This study explores the decision-making processes involved in trajectory extrapolation tasks. Vertical paths were disturbed in the horizontal direction by noise generated by Random Walks (RDW) or Independent and Identically Distributed (IID) noise. In the Experiment, participants (n = 28) extrapolated if a moving point would end right or left of the screen’s center. We modeled path extrapolation using a three-parameter Random Sampling Model (RSM), and a Multi-Layer Perceptron (MLP). Statistically, for RDW paths, the optimal extrapolation strategy is to use the last visible position, whereas for IID paths, it is the average position of the visible trajectory. This normative approach explained 90% of participant responses for RDW and 71% for IID trajectories. The RSM was developed assuming that extrapolations are influenced by a temporal window of visual perception, and two types of uncertainty: sensory noise influencing 1) trajectory perception, and 2) perception of the screen center. The model predicted the proportion of left-right responses for each path. Parameter optimization minimized the RMSE between the proportion of responses in the participant data set and those predicted by the model (Fig.1). For the best RSM the RMSE between real and simulated responses was 0.13, compared to 0.38 for the normative solution. The MLP was trained to predict participants’ Reaction Times and the side of the extrapolated position for each path. Cross-validation accuracy for predicting responses was 93% for RDW and 75% for IID. The RMSE for RTs was 0.13s for IID and 0.21s for RDW. Further validation involved testing the MLP on paths corrupted by noise equivalent to that used for the best-fitting RSM to simulate the sensory input available to participants. The RMSE between the proportion of normative responses for the subject set and those predicted by the MLP on corrupted paths was 0.1. These results suggest that both the RSM and the MLP are effective in capturing the decision-making processes in trajectory extrapolation tasks. The RSM provides a theoretical framework that accounts for uncertainty in participant responses, while the MLP model demonstrates practical efficacy in predicting responses and RT, even when participants deviate from normative decisions. Integrating sensory noise parameters in MLP model testing confirms its robustness and adaptability, making it valuable for understanding and predicting human decision-making in dynamic, uncertain environments.