Hi everyone, I'm working on skin cancer classification and I've extracted the feature from three pre-trained CNN models and concatenated all the features. Finally a dense layer with softmax function was used for classification.
But I encountered constant validation accuracy with lower training accuracy.
How can I solve this problem please. Despite this, I have tried many optimizer functions with different learning rate, regularizer, early_stopping, dropout, and also I used image augmentation
Check for NaN (Not a Number) values in your data and remove them.
Abdelhafid Benchikh
Thank you for your answer. Actually, When I used [dropna()] to remove the NaN data, an error came up:
ValueError: Data cardinality is ambiguous:
x sizes: 9028
y sizes: 476
Make sure all arrays contain the same number of samples.
Maybe this answer help you a little , , addressing the challenge of constant validation accuracy and lower training accuracy in your skin cancer classification project requires a systematic approach of experimentation and optimization. By carefully considering data quality, preprocessing, architecture, hyperparameters, regularization, and various techniques, you can work towards improving your model's performance. Be prepared to iterate and fine-tune your strategies based on the results you observe, and keep a keen eye on the interplay between different factors to achieve the best possible outcome for your classification task
- Badges
- Science topic
- Similar topics
- Data Mining
- Classification