I am trying to improve my scores on leader board and I feel a bit stuck with feature engineering. Does anyone have tips on what worked well for them? I'm especially interested in effective encoding methods or time-series features.
What often helps is stepping back and focusing on simpler, more meaningful features before adding complexity—things like target-aware encodings, rolling statistics, and lag-based time-series features can make a big difference. It’s a bit like Ragdoll Hit: refining control and precision usually gets better results than adding more force. Careful iteration and validation are what ultimately push scores higher.
try target encoding or frequencey encoding
What often helps is stepping back and focusing on simpler, more meaningful features before adding complexity—things like target-aware encodings, rolling statistics, and lag-based time-series features can make a big difference. It’s a bit like Ragdoll Hit: refining control and precision usually gets better results than adding more force. Careful iteration and validation are what ultimately push scores higher.