Interesting with that

The purpose of applying PCA is to extract the informative information of a big data into the first few components. For that PC are arranged according their higher variance. The scatter plot of the attached file denotes how the classes are separated for classification problems. If you wants more separable components then you can apply feature selection approaches after PCA transformation.

I believe you need more than 2 features to separate your classes. By the way, you can try Autoencoder instead of pca which might provides you better features.

Thanks, Mohammad Hossein. I have already tried other methods like Autoencoder. But unfortunately, none of them works well.