Diseases in potato plants can have serious impacts on crop yield and overall plant health, threatening the sustainability of agricultural production. To enhance the accuracy and efficiency of potato leaf disease detection, this study proposes a new approach utilizing Residual Network architecture, a promising technique in image analysis. The dataset used is sourced from the public Kaggle repository, providing the necessary diversity to effectively train the model. The process of splitting the dataset into training and testing data is essential to optimize the performance of the Residual Network algorithm, ensuring that the model can generalize disease patterns well. The research findings highlight that Scenario 1, which adopts a training-to-testing data ratio of 90:10, emerges as the most optimal choice. In testing, this scenario demonstrated superior performance compared to other scenarios, achieving the highest accuracy rate of 76%, indicating its promising performance in classifying data with high accuracy. These results suggest significant potential for this approach in practical applications for effective and efficient potato leaf disease detection, thereby enhancing agricultural productivity and ensuring food security.