I am a beginner in machine learning. I am working on machine learning project. The dataset is not linearly separable (non-linear relationship among the features) and the dataset has 850 features and 7790 instances. I decided to apply dimensionality reduction techniques like kernel PCA or UMAP. After that, I will apply supervised machine-learning algorithm to build a prediction model.
When I use dimensionality reduction, should I calculate the correlation coefficient for non linear relationship?
The answer is certainly yes. In studding a basic concept of PCA related to your query, I suggest you follow https://au.mathworks.com/matlabcentral/answers/445906-pca-in-matlab-reduce-dimensionality and https://au.mathworks.com/help/stats/pca.html
Hey Joynal Abdin,
To start with, you need first to understand the fundamentals in 2 ways:
1. The nature of ML you are going to be working with, the dataset can be converted to the format that is possible to used regardless of its not being linear.
2. Some dimensionality reduction approaches MUST be selected depending on the nature of algorithm and what you are aiming at in terms of post-prediction.
Take a look at these links, there will give you some tips if any:
https://www.nature.com/articles/s41467-022-32243-x
Article Application of variable selection and dimension reduction on...
Article Non-Linear Dimensionality Reduction: A Comparative Performan...
Article A REVIEW ON LINEAR AND NON-LINEAR DIMENSIONALITY REDUCTION TECHNIQUES
To answer this question of "When I use dimensionality reduction, should I calculate the correlation coefficient for non linear relationship?"
This blog will avail some tips too but it also depends, calculation are mainly done still depending on the need:
https://medium.com/analytics-vidhya/how-to-measure-the-non-linear-correlation-between-multiple-variables-804d896760b8
https://arxiv.org/pdf/0901.0537
https://www.biorxiv.org/content/10.1101/273862v2.full
Hope it helps
- Badges
- Science topic