# Neural Network Architecture - Accuracy issue

**URL:** <https://discuss.ai.google.dev/t/neural-network-architecture-accuracy-issue/29562>\
**Category:** Keras\
**Tags:** datasets, help\_request\
**Created:** [June 2, 2022, 11:25am UTC](https://discuss.ai.google.dev/t/neural-network-architecture-accuracy-issue/29562 "2022-06-02T11:25:46Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![mtalhaubaid](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/mtalhaubaid/32/11233_2.png) [@mtalhaubaid](https://discuss.ai.google.dev/u/mtalhaubaid)\
**Post date:** [June 2, 2022, 11:25am UTC](https://discuss.ai.google.dev/t/neural-network-architecture-accuracy-issue/29562/1 "2022-06-02T11:25:47Z")

</div>

I have a structured dataset having features and 570 classes. I have done a lot of experiments with the architecture but accuracy doesn’t increase. I also normalized the dataset before passing it into the model.  
Here is the model architecture.

loss=‘categorical\_crossentropy’,  
optimizer=‘adam’ also tried SGD(lr=0.01) as well

md=model.fit(X\_train,Y\_train,validation\_data=(X\_test,Y\_test\_c),epochs=1500,batch\_size=512, verbose=2 ,shuffle=True, callbacks=[es])

 ![image](https://d3qe71uytubmmx.cloudfront.net/original/2X/2/2f1d8f825ab575bdf6ff3107aaf9878e1ada14e5.png)

---

<div class="post-metadata">

**Author:** ![innat](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/innat/32/13542_2.png) [@innat](https://discuss.ai.google.dev/u/innat)\
**Post date:** [June 2, 2022, 7:56pm UTC](https://discuss.ai.google.dev/t/neural-network-architecture-accuracy-issue/29562/3 "2022-06-02T19:56:10Z")

</div>

It’s ambiguous what you’re doing.  
Have you tried TabNet?

---

<div class="post-metadata">

**Author:** ![mtalhaubaid](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/mtalhaubaid/32/11233_2.png) [@mtalhaubaid](https://discuss.ai.google.dev/u/mtalhaubaid)\
**Post date:** [June 3, 2022, 4:39pm UTC](https://discuss.ai.google.dev/t/neural-network-architecture-accuracy-issue/29562/4 "2022-06-03T16:39:02Z")

</div>

No, I haven’t. I try to make my own architecture