Glitches, non-Gaussian transient waves which mimic gravitational-wave signals, are abundant in detectors and impact data quality. Therefore, identifying glitch type and eliminating them is crucial to unveil true astrophysical events. In this study, we evaluate the performance of logistic regression, extreme gradient boost, and support vector machines for glitch classification using features extracted via transfer learning on Inception-v3 and ResNet-50 models. We used two transfer learning strategies: fine-tuning pre-trained models with our dataset and using pre-trained models as feature extractors. Our results show that transfer learning significantly reduces training time compared to fine-tuning. The transfer learning method achieved a classification accuracy of \(93.98\%\) with the lowest training time of 37.6 s. Transfer learning resulted in 24 and 31 times faster training for ResNet-50 and Inception-v3, respectively, proving highly beneficial for glitch classification in the LIGO experiment.