Hello,
I’m creating a model with 3 convolutional and maxpool layers. Convolutional filter size is (3, 3) and filter starting from 16 then 32 and finally 64. All the maxpool layers have (2, 2) filter.
Finally two dense layers respectively 512 and 1 neurons. Using binary cross entropy as loss and Adam as optimizer with learning rate 0.01.
But my model accuracy is only 50%. Please suggest me how can I achieve higher accuracy.
Hey @MD_Maruf_Al_Hossain,
Well there are many options you can consider to get higher accuracy it all needs experiment until you find best model but i will give you some factors to think about and give it a try to play with it.
Data Augmentation : Increase the diversity of your training data by applying data augmentation techniques such as rotation, translation, flipping, and zooming. This can help the model generalize better.
Dropout : Add dropout layers in between dense layers to prevent overfitting. Dropout helps to regularize the model and improve its generalization as well.
Learning Rate : Experiment with different learning rates. A learning rate of 0.01 might be too high for your model. Try reducing it and see if the model converges better. Learning rate scheduling (e.g., reducing the learning rate over time) can also be helpful.
Batch Normalization : Add batch normalization layers after convolutional and dense layers. Batch normalization can improve the convergence speed and overall performance of your model.
Hyperparameter Tuning : Experiment with different hyperparameters like the number of filters, kernel size, and architecture. You can use techniques like grid search or random search to find the best combination.
Normalization : Ensure that your input data is properly normalized (e.g., mean-centered and scaled). Normalization can help the model converge faster.
It’s just a brief of options you can think about but first thing you need to consider is “The more data the better” so start by augmenting your data and then start to consider other factors.
I hope it helps you and feel free to ask for more clarifications.
Regards,
Jamal
