[tesseract-ocr] Finetuning Tesseract to detect numbers

I'm trying to finetune Tesseract to recognize digits only but I'm not getting good results so far. I continued the training from Arabic language "ara" since the digits I'm trying to recognize are Arabic numbers. The training will stop early at 0.01 error rate but the results on testing data is

Re: [tesseract-ocr] Help with detecting a code on an image

*Thanks for all the help.* *I tried closing as a morphological transformation and also used a threshold to get a black and white picture.* *I got "BYC3 .- A4UFYY.QLBGY -+ " as a answer this is almost right.* *Is there anything else that i can do to improve the detection?* *i attached the python