Renal cancer is among the most prevalent types of cancer, affecting a large number of individuals worldwide. In order to enable patients to receive treatment as early as possible and increase the chance of survival, it is necessary to diagnose patients quickly. As a critical foundation for disease diagnosis, histopathological images play a pivotal role in cancer diagnosis by offering detailed visual representations of microscopic cellular structures, including markers and surface features essential for assessing disease progression. However, this requires a pathologist with extensive experience to make a judgment, which requires a long time, cost, and a large amount of time and labor costs. In response, the emergence of computer-assisted diagnosis systems presents a promising solution to this challenge. With the continuous development and progress of deep learning in recent years, utilizing more advanced deep neural networks enables us to achieve improved classification outcomes. This study uses ResNet50 to extract and learn deep features of histopathological images, selects features through the L1 norm, and then uses the MLP-RF classifier to classify healthy and renal cancer cells based on the selected features. Unlike previous classification methods that rely solely on deep learning or machine learning, the MLP-RF classifier addresses the limitations inherent in each individual approach by combining their strengths. The findings of the study demonstrate the above method allows us to achieve better classification results on the used dataset.