regardless of the results, it is good practices to let your research gets evaluated by a steer committees in a specialized conference.

Are you doing transfer learning? If not, then your data set is orders of magnitude too small.

It is not sufficient. If you are introducing new method, it is okay, however., getting large datasets of images are freely available, so you may try it e.g., https://www.robots.ox.ac.uk/~vgg/data/

http://homepages.inf.ed.ac.uk/rbf/CVonline/Imagedbase.htm

Research cannot be evaluated by accuracy values only. If you think you have improved upon the recognition algorithm as compared to earlier works, in a substantial way, then you may submit highlighting your contributions through comparative analysis.

More data may not be a criteria for improving accuracy. However, for stability it is necessary to go for large number of inputs. You have to think of improving the method or model.

Also, provide a comparative analysis with existing similar methods. This will provide your method a standing.

What kind of data you're using? So, I may help you to find out more data for your classification. In my point of view, the accuracy you've got is enough for some models.

Bad results are publishable as long as your paper introduces some novelty, so as long as you have made a contribution the result might be publishable.

This said you need to make sure to discuss your experiments and results!

It is perfectly fine to submit your research paper with the level of results you have got. You can create a proper justification for the same.

Sure. As long as your work defines a novel approach and has not been explored before, or partially explored, it is fine to document your work in shape of a research paper. I primarily work on NLP, and I have seen top-notch conference papers with 56-65% validation based accuracy which made the cut for acceptance and thereafter publication.

Yes submitting your paper will help you get some feedbacks. Furthermore, I think an accuracy of 80% is quite good. So submit your paper and be ready to revise it if it gets through the first acceptation step (resubmit with major revisions).

Good luck!

It doesn't matter what accuracy measures you have. The idea is how much enhancement your technique achieved when compared with the state of the art techniques on the same data you got.

If you are introducing a new technique, you need to test it on different datasets, with different characteristics, and from different sources to avoid overfitting.

Additionally, Accuracy is not always an indicator of the superiority of one technique on another. You may need to deploy other evaluation metrics such as AUC, Geometric Mean, etc.

Research like Repath something invite new one compare to old one also true research never failure 80 % data true means it's ok not accurate like simulation Ok

Ok to submit and have good feedback. Then, reflect the feedback to further improvements of the work. 

Dear Anonymous researcher,

A true research doesn't mean that always you will get better result on your model. To highlight the draw backs of the model with a proper explanation is also included in research.

Similarly, efficiency of your model never measured with the amount of data you have used. How much data you have used in your model, with this data if your model obtain better result over other existing models in same domain then you can go for your model.

Thanks

Works which cannot be improved further by a researcher can be termed as the best he/she can do. If the results can be established to be poor basing upon certain criteria then always those can be improved.

Publication for the sake of publishing is definitely not advisable. One has to see the addition it makes to the existing knowledge.

Even minor improvement can be considered for publication. If the author himself feels that certain results are poor then how can he expect those results to be used or referred by anyone else?