A Second Approach To Automating Detection of "Random-Looking" Domain Names: Neural Networks/Deep Learning

This technical blog demonstrates building and evaluating a Keras-based neural network (embedding layer + Conv1D, pooling, and dense layers) to detect random-looking/algorithmic domain names (DGAs) in Farsight Channel 204 DNS telemetry. It details preprocessing (character-tokenization, padding/trimming to 20 chars), categorical encoding vs embeddings, oversampling and class-weighting to address extreme class imbalance, training on millions of records, and validation results showing a small fraction of domains flagged as random-looking; it is a methodological report rather than an incident notification.