The field of gravitational wave astronomy has made remarkable progress in recent years, with 90 successful detections by Advanced LIGO and Advanced Virgo in three observing runs. The use of deep learning to analyze gravitational wave data is an active area of research with the potential to improve our ability to detect and study these signals. However, the inherent black-box nature of deep learning models poses challenges in interpreting their predictions. To address this, we applied gradient-weighted class activation mapping technique to visualize our 4-class classification model trained on signals from binary black hole mergers, neutron star-black hole mergers, binary neutron star mergers, and noise. The visualization allows us to gain insight into which part of the strain was most influential in the model's predictions. The visualized maps indicated that as the signal duration increased, the model prioritized data before the merger time.