Histopathological oral images are one of the popular methodologies to diagnose oral squamous cell carcinoma. Computerized diagnostic systems need to be developed in order to process huge number of histopathological oral images for each subject and helping the clinician in classifying the oral images as either normal class or oral squamous cell carcinoma class. This research work uses deep transfer learning approaches namely MobileNetV3, InceptionV2, and EfficientNetB3 for extracting the features from histopathological oral images. Spotted Hyena Optimizer is used as feature transform for improving classification accuracy. Two dense layers are used for classifying the transformed features. 2784 & 3632 number of histopathological oral images are considered in this study under the two classes as normal and oral squamous cell carcinoma respectively. With the proposed spotted Hyena Optimizer, the maximum accuracy attained is 93% for the features extracted through InceptionV2 model. Relatively, the highest accuracy attained is 89% if the optimizer is not used for the features extracted through MobileNetV3 model.