Representation learning significantly enhances botnet attack detection capabilities by transforming raw network traffic data into structured representations. This paper focuses on leveraging deep learning techniques, particularly ResNet-18 convolutional neural networks (CNNs), to detect botnet activities. Using the CTU-13 dataset, we demonstrate how the MinMaxScaler can preprocess the data, enabling effective feature extraction. By identifying subtle anomalies associated with botnet behavior, our approach enhanced detection accuracy while reducing false positives. Integrating representation learning with traditional machine learning classifiers further boosts performance, making it an essential strategy for developing advanced cybersecurity solutions against evolving botnet threats.