Machine Learning Captcha Project

For this machine learning course, we assessed the mathematical efficiency and eventually implemented codes to teach a computer to solve difficult problems designed to stump robots and computers. This specific project assessed Captchas, which are used to make sure robots and computers can't mine data in secure sites.

In preparation for this project, we were given a framework that basically allowed for the computer to read the data, which was comprised of examples of captchas (made up of six characters {numbers or letters] in total). I will provide the results in detail below, showing the rates of success with each character. As you might expect, the characters on the edges are the easiest to get, but a minimum character accuracy of 82.4% on character 4, was a solid result.  This code was completed within Visual Studio Code using Python as the programming language.

val_Char_1_accuracy: 0.9080 - val_Char_2_accuracy: 0.8440 - val_Char_3_accuracy: 0.8540 - val_Char_4_accuracy: 0.8520 - val_Char_5_accuracy: 0.8240 - val_Char_6_accuracy: 0.9200 

Below is the code that I changed. The McDigit base predicted at about a 30-35% rate per character, I believe. Below that is the image of what the Captcha data went through in order to become more "readable" for the computer.