This paper aims to verify if the seismic slowness log estimated through the supervised machine learning K-nearest neighbor (KNN) algorithm can be a feasible alternative to replace the sonic well log as input for the seismic well tie in a dataset from the Recôncavo Basin. The training and optimization of the regression were performed in a dataset composed of 17 well logs with petrophysical information of gamma-ray, deep and shallow resistivities, and the geological formation, e.g, Pojuca, Marfim, Maracangalha, Candeias, São Sebastião, Água Grande, and Sergi Formations. The metric to evaluate the regressions was the mean absolute error of the measured property and the prediction. The holdout cross-validation technique was applied to avoid overfitting, and a well log was separated as a blind test to verify the prediction in an unknown dataset. Furthermore, synthetic seismic traces were generated from the slowness log and the prediction using the KNN. The comparison between them shows outstanding results in the visual analysis of the peaks and amplitudes of the main seismic events. In addition, the comparison between the seismic traces close to the synthetic seismic traces reveals a better correlation to the calculated traces using the slowness predicted by the KNN algorithm.