profanity-check

Achieves 95% test accuracy and balanced metrics like precision and recall, outperforming similar libraries such as profanity-filter, as shown in the README's comparison table.

Processes predictions 300–4000 times faster than profanity-filter, with benchmarks showing 0.2 ms per prediction, making it ideal for real-time applications like chat moderation.

Uses a machine learning model without explicit blacklists, enabling it to detect phrases like 'You cocksucker' that wordlist-based libraries miss, as explained in the No Explicit Blacklist section.

Provides predict_prob() to output the probability a string is offensive, offering nuanced insights for fine-grained moderation decisions, as demonstrated in the usage examples.