The liver is one of the most important parts of human life. Nowadays, everywhere on the globe, people’s lives are affected by the malfunctioning of the liver organ. Due to many aspects of the non-proper day-to-day functions of the organs, such as fatty liver, cirrhosis, and hepatitis. Many humans' irregular diets, alcohol consumption, etc. are leading them to get affected by liver diseases. The mortality rate of the affected patients is high. Identifying the affected patients in their earlier stages by using common tests like biopsy, X-ray, CT scans, and MRI scans will help the physician do the proper treatments to recover from the deadly disease. In this research paper, we make use of medical MR scans of the subjects to apply three classification techniques to identify if the liver is healthy or impacted by illness. Also suggests a novel strategy that uses the Social Spider Algorithm (SSA) to reduce the features of MR images of the medical dataset. After feature reduction, the well-known CNN architectures GoogleNet, ResNet50, and VGG16 are used for liver image classification. The decreased feature set from the MR images that have been preprocessed and the SSA reduction technique are used to train and fine-tune the CNN models. Using a publicly available dataset of liver image data, the accuracy, precision, recall, sensitivity, specificity, and F1Score of 50 epochs with batch size-32 of each of the proposed CNN architectures are evaluated with and without SSA. The paper contributes to enhancing early liver disease detection by integrating the SSA for MRI feature reduction and exploring optimal CNN –GoogleNet-architectures for improved classification accuracy attained score is 99.41%.