Adbot Ad Engagement Forecasting Challenge
'%3e%3cpath fill='%23001489' d='M0 0v6h9V0z'/%3e%3cpath fill='%23e03c31' d='M0 0v3h9V0z'/%3e%3cg stroke='%23fff' stroke-width='2'%3e%3cpath id='d' d='m0 0 4.5 3L0 6m4.5-3H9'/%3e%3cuse xlink:href='%23b' stroke='%23ffb81c' clip-path='url(%23c)'/%3e%3c/g%3e%3cuse xlink:href='%23d' fill='none' stroke='%23007749' stroke-width='1.2'/%3e%3c/g%3e%3c/svg%3e)
Helping South Africa
$500 USD
Completed (almost 2 years ago)
Forecast
452 joined
113 active
Start
Apr 04, 24
Close
May 19, 24
Reveal
May 19, 24
feature importance
Help Ā· 13 May 2024, 15:10 Ā· 10
In this challenge, we are tasked with two objectives: first, analyzing how features contribute to the target, and second, forecasting the targets. I would like to know if we need to use the same model for both tasks, or if we can use two different models—one to study feature importance on the targets in the training dataset, and another to forecast the clicks using only the time series data of the clicks.
is this approache allowed?
I was wondering the same. Given that they do not provide the features for the target dates, it became a forecasting challenge and those features are relatively useless
I assumed that we could use random forest for feature importance as an example then use the top 5 features to implement a multi-step forecasting model using the five features which can be an LSTM or tree based model.
that would be one way to do it, but what if not using any features at all will give a better score
wait are you using no feature at all😅
yes
16.41180069 with no features??😲oh you magician!!!!
Which forecasting model are u using arima and how or any machine learning model for forecasting,can u give any hints
i am using catboost with a function f(y) for the final prediction. It was mostly luck finding f(y) that would reduce the rmse very well, and i am not sure as i may be overfitting to the 20% so there's that.
How did u created the feature for the test dataset did u used the the forecasted values for the coast , impression or feature engineered new columns like aggregation of the clicks for model
The description suggests that you have to use the same model for both tasks.