Högskolan i Skövde

During the last decade, there has been an increase in the number of automated programs (bots) that perform tasks such as harvesting information or making posts on social media. CAPTCHA was developed as a defense against bots, but several common CAPTCHAs have usability issues and are difficult for users to solve. This project aims to determine if a non-visible user input-based CAPTCHA can help solve this problem. The CAPTCHA looks for patterns in the user input, that is, signs that the input is controlled by scripted logic.

The CAPTCHA is evaluated by looking at how capable it is at identifying patterns, human mouse movements and bot-controlled mouse movements. Additionally, it is investigated if there exists a data sequence size at which the pattern recognition algorithm can most successfully detect patterns and avoid false negatives. The results showed that interval sizes 40-50 provide the best results. Using these sizes, the pattern recognition algorithm was able to fulfill the commonly accepted rates of at least a 99 % success rate and at most a 10 % false negative rate. This shows that the CAPTCHA is robust under the circumstances investigated.