0
$\begingroup$

I have been working on a project with low features and only few entry fields ( 4 to be exact ). All the data in the dataset is barely correlated to each other. Is there some organized way or approaches to tackle such data and thus build a predictive model out of it. To give you a little glimpse of the dataset, here are a few columns from the

+----+----------+-----------+---------+
| ID | OAG CODE | CONSIGNEE | SHIPPER |
+----+----------+-----------+---------+
| 11 | 665      | 1001      | 20100   |
| 11 | 665      | 1006      | 20105   |
| 13 | 667      | 1023      | 20110   |
| 13 | 669      | 1015      | 20104   |
| 13 | 669      | 1006      | 20105   |
+----+----------+-----------+---------+

I want to perform EDA over the data and also build a predictive model from the same. Please point out some of the standard techniques and methods for tackling such a problem.

$\endgroup$
2
  • $\begingroup$ The values you have provided all look like identity columns. Is there anything in the data that makes sense to you intuitively? What would you like to predict from this data? $\endgroup$
    – Skiddles
    Nov 30 '18 at 17:32
  • $\begingroup$ @Skiddles, I have to predict if given a OAG code, what all SHIPPER or CONSIGNEE will he be shipping his code to. $\endgroup$
    – thanatoz
    Dec 1 '18 at 10:43
2
$\begingroup$

It seems like a challenging problem. If it were my task, I would start with a probabilistic approach like apriori, but you may want to check out Naive Bayes based approach. There are some differences in these approaches, but, either one may produce decent results. More generally, I think the analysis you want to perform is $Association\ Rule\ Learning$.

One thing you want to be aware of with this data is that even though it is all numeric, you cannot treat the data as numeric data. You should treat each element as textual data because there is no inherent relationship from one consignee number to the next for example. This fact also changes how you can approach EDA. For example, there is no value to knowing the mean or the std dev of the OAG Code nor is there a linear relationship between OAG Code and Consignee code, that is, "as OAG Code increases, so does Consignee code."

HTH

$\endgroup$
2
$\begingroup$

It is really hard to answer your question because there is too little information. Try to do some EDA and attach it to your question because EDA defines the model.

Anyway, for weakly correlated data, try to use k-NN. If you use Python, scikit-learn has an implementation. Also, you can try decision trees.

Sorry, but I am also new in DS and can be wrong:)

$\endgroup$

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