# Handwritten Digits Performance

**URL:** <https://community.deeplearning.ai/t/handwritten-digits-performance/706656>\
**Category:** AI Discussions\
**Tags:** tensorflow, computer-vision, project\
**Created:** [October 11, 2024, 10:42am UTC](https://community.deeplearning.ai/t/handwritten-digits-performance/706656 "2024-10-11T10:42:38Z")\
**Posts on this page:** 5\
**Page:** 1

<div class="post-metadata">

**Author:** ![Godugu\_Anil\_Himam](https://avatars.discourse-cdn.com/v4/letter/g/9dc877/32.png) [@Godugu\_Anil\_Himam](https://community.deeplearning.ai/u/Godugu_Anil_Himam)\
**Post date:** [October 11, 2024, 10:42am UTC](https://community.deeplearning.ai/t/handwritten-digits-performance/706656/1 "2024-10-11T10:42:38Z")

</div>

Hey everyone,

I have been working on a project to recognise hand written digits for the Digit Recognizer competition on Kaggle. I have been able to consistently develop the performance of the model from 95% to 99% on the public score. I have now stumbled upon a block and am not able to increase the performance of the model any further. The current performance of the model is as follows:

- Training Set: 100% Accuracy; 3% Loss
- Validation Set: 99.4% Accuracy; 5% Loss
- Testing Set: 99.19% Accuracy; 6.5% Loss

Kindly help me out by providing tips, advice or new approaches to increase the performance model. Here is a link to the notebook [My Notebook](https://www.kaggle.com/code/goduguanilhimam/soon-to-be-god-of-handwritten-digits)

Thank you so much for helping me out 🙂 🙃

---

<div class="post-metadata">

**Author:** ![TMosh](https://sea1.discourse-cdn.com/dlai/user_avatar/community.deeplearning.ai/tmosh/32/9321_2.png) [@TMosh](https://community.deeplearning.ai/u/TMosh)\
**Post date:** [October 11, 2024, 3:02pm UTC](https://community.deeplearning.ai/t/handwritten-digits-performance/706656/2 "2024-10-11T15:02:14Z")

</div>

Perhaps you have reached the useful limit.

---

<div class="post-metadata">

**Author:** ![Nevermnd](https://sea1.discourse-cdn.com/dlai/user_avatar/community.deeplearning.ai/nevermnd/32/126021_2.png) [@Nevermnd](https://community.deeplearning.ai/u/Nevermnd)\
**Post date:** [October 11, 2024, 3:59pm UTC](https://community.deeplearning.ai/t/handwritten-digits-performance/706656/3 "2024-10-11T15:59:35Z")

</div>

@TMosh @Godugu_Anil_Himam I recall Prof. Ng’s, what was it, course three (?) of DLS and at least it sat me down to say hmmm…

---

<div class="post-metadata">

**Author:** ![TMosh](https://sea1.discourse-cdn.com/dlai/user_avatar/community.deeplearning.ai/tmosh/32/9321_2.png) [@TMosh](https://community.deeplearning.ai/u/TMosh)\
**Post date:** [October 11, 2024, 4:33pm UTC](https://community.deeplearning.ai/t/handwritten-digits-performance/706656/4 "2024-10-11T16:33:20Z")

</div>

You don’t really need to tag individuals in your reply, because “Watching” is automatically set when anyone replies on a thread.

---

<div class="post-metadata">

**Author:** ![Nevermnd](https://sea1.discourse-cdn.com/dlai/user_avatar/community.deeplearning.ai/nevermnd/32/126021_2.png) [@Nevermnd](https://community.deeplearning.ai/u/Nevermnd)\
**Post date:** [October 11, 2024, 5:56pm UTC](https://community.deeplearning.ai/t/handwritten-digits-performance/706656/5 "2024-10-11T17:56:08Z")

</div>

Sorry, only trying to help…