This data was tallied by using Machine Learning and by using volunteers, by identifying handwritten texts from the IEBC forms. These forms can be publically obtained form the following link: https://forms.iebc.or.ke

Unfortunately the method used is fairly inaccurate. It should not be relied on. The purpose of this project was to assess the performance of current OCR technologies.

However, it is important to note that the models used for this project were not trained for along time. Mostly due to limited labelled data. Once the official tally is complete, we can assess the performance of the models and improve on it.,

Its a fairly long process

1. The data is first obtained from the official website in PDF form
2. The PDFs are the converted to images for easier processing
3. The images are then cropped using a CNN deep neural network. Specifically the section with the handwritten tally.
4. The handwritten tally is then fed to an OCR model for text extraction.
5. The extracted text is then parsed and the numbers are extracted.
6. The numbers are then fed to Database, which is then relayed here

Please see the data set below for more information

Yes you can. Please refer to this github repository The data is made available as CSV. Note that this is not the official data, but crowdsourced from the public.