Using only one aggregation (i.e. mean) for Nighttime lights model features instead of all the aggregation variations

[joshuacortez]

Among the poverty mapping model features, we're using many different versions of the nighttime lights features:

• avg_rad_mean
• avg_rad_median
• avg_rad_min
• avg_rad_max
• avg_rad_std

I suspect that these features are very highly correlated. Having highly correlated features can distort the feature importance and possibly bring down the accuracy. When we have the time we could try a model experiment where we only keep one of these features for the model