Mtoto News Childline Kenya Call Volume Prediction Challenge
2000 Zindi Points
Help Kenya's child protection hotline forecast how many calls they will receive each hour each day
11 April–4 August 2019 23:59
191 data scientists enrolled, 38 on the leaderboard
No test set and submission file is not where clear
published 22 May 2019, 10:07

we dont have test set and after prediction how will the submission file look like after modelling am having RMSE = 18 and submitting is giving me 50 on the scoreboard

Hi @Nasere,

"we dont have test set" : this is a forecasting competition a.k.a you will use the data in certain period to predict future values in another period. In other words, there is no test set as in classical competitions.

"and after prediction how will the submission file look like": The submission format is describe in the section Data, have a look! If you got score this means that the file you submitted is ok.

"after modelling am having RMSE = 18 and submitting is giving me 50 on the scoreboard":

1. How did you train you model ? With time series you could not randomly split your data into trainning, and test set! For example if you are using Python, don't use the function train_test_split of scikit-learn. Instead, split by period, for example the data from january to april will be your trainning set, and the remaining the test set or use the function TimeSeriesSplit of scikit-learn.

2. Check if you are not overfitting!

3. Check if your predictions index match with the submission file one.

Have fun!

Thanks for the insights on how to go about splitting time series data.