1

I am newbie in amazon Machine Learning. I could able to generate prediction using Machine learning using "linear regression", however i could not able to find suitable example "For this product, how many units will sell?". My expectation would be to find how many products required in each category in future using prediction. I was expecting simply X & Y axis can be defined in input however amazon Machine learning is expecting only targets and the graph automatically generated. X axis always takes number of records from the input CSV instead count of the product sold.

I need guidance how to proceed further or document or materials to refer further.

Thanks, Raj

John Rotenstein
  • 241,921
  • 22
  • 380
  • 470
user8432615
  • 33
  • 2

1 Answers1

2

Amazon Machine Learning works by supervised training. You must provide it with historical data (as much as possible) and a result. It can then predict the result when given a set of inputs.

For example, if you were predicting sales for Ice Cream, you might provide input data such as:

  • Month
  • Day of Week (Mon-Sun)
  • Weather (Sunny, Windy, Hot, Cold)
  • Is it a public holiday?
  • Is it a school holiday?
  • Temperature

You would also provide a result, such as the number of ice creams sold.

You take as much of the above data as possible and upload it to Amazon Machine Learning, which then generates a model that best fits your data. Internally, it will use linear regression but it will automatically choose the model using many different predictive algorithms.

Then, you can ask it to predict values for a given set of inputs, such as:

  • July
  • Wednesday
  • Windy day
  • Not public holiday
  • School holiday
  • 50 F temperature

It will then generate a predicted level of Ice Cream sales.

Machine Learning works best if you can provide it with many examples so that it can better build the predictive algorithm. You should also provide many different variables. For example, don't just give it a date, because the date is different every day. Instead, break it down into Day of Week, Month, Quarter, Public Holiday -- so that it can find deeper meaning in the day itself rather than just being the 128th day of the year.

For further details, see: Amazon Machine Learning Key Concepts

John Rotenstein
  • 241,921
  • 22
  • 380
  • 470