Deep Brain Stimulation (DBS) for Parkinson’s disease has been identified as an effective treatment, especially focusing on the subthalamic nucleus (STN). The placement of electrodes is essential for the success of DBS in treating motor symptoms of Parkinson’s disease. One of the main challenges in DBS surgery is the precise targeting within the STN, a region in the basal ganglia. The dorsolateral region is considered the relevant site for electrode placement among STN subregions due to its association with improved motor outcomes. However, localizing this region intraoperatively remains challenging due to inter-patient variability and the absence of consistent anatomical markers. In the current study, we propose an unsupervised Attention-based Contrastive Variational Autoencoder (ACVAE) model to identify the dorsolateral STN using each patient’s microelectrode recording (MER) data of different depths. We calculated Power Spectral Density (PSD) and spectrograms from Beta-band power spectrum for different depths for each individual. The proposed ACVAE model employs a dual-encoder architecture: one encoder learns depth-specific local patterns from PSD and spectrogram data across multiple depths, and the other encoder captures patient-level global context. Attention layers integrated into both encoders dynamically highlight the most informative neural segments, enhancing the model’s sensitivity. To validate the results, we also identified the dorsolateral region using temporal and spectral features, including spike rate, beta-band power, and other parameters. We cross-validated results identified by the proposed model with results identified using the conventional approach. The proposed model achieved an average accuracy of 90% with an F1-score of 0.92. We also performed leave-one-out cross-validation to assess the results and achieved an average accuracy of 88% with an F1 Score of 0.89. This demonstrates the model’s robustness in generalizing across individuals without requiring labeled clinical outcomes. Our approach offers a data-driven and patient-specific method for refining electrode placement strategies in DBS, potentially enhancing intraoperative decision-making and clinical outcomes.