Parkinson’s disease (PD) is a neurological disorder with a rising global prevalence. A number of surgical approaches are used to treat PD, among which Deep Brain Stimulation (DBS) is a widely chosen treatment to alleviate PD motor symptoms. DBS involves the implantation of microelectrodes in target brain regions, most commonly the subthalamic nucleus (STN). Accurate localization of the STN is essential for maximizing therapeutic outcomes and minimizing surgical risk. This study addresses the challenge of STN localization using machine learning (ML) techniques applied to microelectrode recordings (MER) collected intraoperatively. A dataset of 36 electrode trajectories from bilateral DBS procedures in seven patients was analyzed. Eighteen time-domain features were extracted from the MER signals. The goal was to develop a classification model capable of distinguishing whether a recording originated within the STN or an adjacent region. Five machine learning (ML) algorithms were evaluated: Random Forest (RF), Gradient Boosting Machine (GBM), Support Vector Machines with linear and radial basis function kernels (SVM-L, SVM-RBF), and k-Nearest Neighbors (k-NN). RF achieved the highest performance, with 73% sensitivity, 86% specificity, and an AUC-ROC of 91%. GBM showed comparable results: 77% sensitivity, 86% specificity, and an AUC-ROC of 90%. The k-NN model also performed well, with an AUC-ROC of 89%, while the SVM models achieved AUC-ROC values of 88% (linear kernel) and 86% (RBF kernel). Feature selection analysis indicated that threshold, baseline amplitude value, and root mean square (RMS) are key variables for delineating the STN. These findings demonstrate that ML-based analysis of MER data can support an objective localization of the STN in real-time DBS procedures, enhancing surgical accuracy and patient safety.