1
$\begingroup$

I have a dataset of over $1000$ instances where the target value is a binary class $0$ or $1$. Each instance has $13$ features. The ultimate goal is to classify a subsequent dataset (same dimension, but this time the target value is unknown).

Before taking any action, I'd like to explore the dataset to maybe catch some interesting trends or behaviour.

Yet, there are $13$ dimensions so I can't just plot the data on the screen.

My first thought was to create several plots, each time with a different pair of attributes, for instance $ (a_1, a_2), (a_1, a_3)... (a_1, a_{13})$ and so on. However, it might be difficult to get any useful information using this representation.

Is there any way I could represent in one plot such a high dimensional set, preferably as a Python library?

$\endgroup$

2 Answers 2

4
$\begingroup$
  1. You can first plot the correlation matrix to get a better feeling of the data.

  2. You could subplot the distributions of all the feature to further explore.

  3. 13 features VS 1000 instances, if these features can characterize your problem well and data are not too noisy, you could get a good model for testing.

  4. If you do think the feature dimension is high, besides looking at correlation matrix, you could do some feature selection. For example, the feature_importance_ method of RandomForestClassifier in Sklearn could help to rank the features.

You can find further details here:

https://www.kaggle.com/ekami66/detailed-exploratory-data-analysis-with-python https://www.datacamp.com/community/tutorials/exploratory-data-analysis-python

$\endgroup$
3
$\begingroup$

$13$ features is not high dimension at all. You can try to plot a correlation matrix. Like this one:

Image 1

In addition, you can try PCA, and map the data into lower dimension, and plot. Here is an example in Python.

$\endgroup$
1
  • $\begingroup$ Thank you for your answer. Regarding the correlation matrix, how can I interpret it? My first assumption would be to prune the features with an overall low correlation with respect to other attributes. Am I correct? $\endgroup$
    – Pierre P.
    Feb 10, 2018 at 21:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.