I have a customer's online data. I have data such as the number of items purchased by the customer, the number and number of keyword queries for the customer, the age of the customer, the residential area of the customer, and the information of the purchased product. 1. I want to forecast the trend of the product. 2. I want to forecast sales volume by main product. However, I have only 6 months of data. How can I solve the problem 1,2?
Asked
Active
Viewed 25 times
2 Answers
0
It depends a lot on how easy/difficult it is to identify trend or autocorrelations within the six months you mention to have data for your product. It is likely that 6 months of data may be too little for applying more complex models.
Short ts make it complicated to properly evaluate the model performance (thinking here about cross-validation for a good performance assessment) - thus, it is challenging to properly select the best model among a number of competing models you may consider.
Usually, simple methods (mean, median, naive) perform better for short series.
Here it is a post that you may find insightful.
Luminita
- 268
- 1
- 5
0
From what you are saying it is very difficult to judge. You could apply a lot of different machine learning methods (for example a Gaussian Process, linear/kernel regression, neural networks...) depending on the structure and the size of your dataset. But what all will have in common, that it will be difficult to predict for "time regions" where you have no data. Here I assume, that the behavior of the costumers is variable over the year (christmas shopping etc.). If you assume that there are no significant changes, the problem of having only 6 months of data would decrease.
Balou
- 19
- 4