A fashion retailer is forecasting weekly online sales to allocate ad budget. Customer response has shifted in the past year because of changing social media trends, new competitors, and a platform algorithm change, so managers want validation to emphasize the most recent data and reduce reliance on older periods. Which time-series cross-validation approach is most appropriate?

Question

BlackTom AI · Accepted Answer

Moving-window cross-validation emphasizes recent observations by training models on a fixed-length sliding window and validating on the immediately following period, so it directly reduces reliance on older data.

1) Relevant concepts
- Nonstationarity / concept drift: when relationships change over time (social trends, competitors, algorithm updates), older data may be misleading.  
- Time-series CV variants: expanding window (growing training set), moving window (fixed-size sliding training set), single holdout (one train/test split), and standard k-fold (random splits ignoring time order).  
- Goal: validate model performance while weighting recent patterns more heavily.

2) Step-by-step reasoning
- Managers want validation that emphasizes recent data and downweights older periods because customer response has shifted.  
- Moving-window CV builds multiple folds where each fold trains on the most recent W periods and validates on the subsequent period(s). Example: with weekly data and W = 12, fold 1 trains on weeks 1–12, validate week 13; fold 2 trains on weeks 2–13, validate week 14; and so on. This produces multiple performance estimates focused on recent windows rather than cumulative history.

3) Intermediate indexing example (illustrative)
- Window length W = 12 weeks.  
- Fold 1: train = weeks [1..12], validate = week 13.  
- Fold 2: train = weeks [2..13], validate = week 14.  
- Continue sliding the window to test robustness across recent periods.

4) Why the correct answer is correct
- Moving-window CV explicitly downweights old information by excluding it from the training window for later folds, matching the stated need to emphasize recent data and validate under shifting conditions.

5) Why other options are incorrect
- Expanding window includes all past data in each fold; it increasingly relies on older periods and is ill-suited for recent drift.  
- Hold out the most recent month only gives a single, noisy performance estimate and no assessment of variability across recent periods.  
- K-fold cross-validation (random folds) breaks temporal order, causing information leakage and invalid time-series evaluation.

6) Conclusion
- Therefore, Time-series cross validation with moving window is the most appropriate choice, which matches the provided correct answer.

类似问题

When evaluating a new project, firms should include in the projected cash flows all of the following EXCEPT:

Provide the best rationale for what's wrong with supply chains today?

What is Blockchain and how is it relevant to supply chain management?

Why aren't most products today recycled and/or reused?

What can be said about Two Tier vs. Three Tier distribution?

更多留学生实用工具

智能学习助手

风格化写作助手

论文查重助手

文献引用助手

课堂转译助手

课堂笔记助手

Quiz搜索助手

学校历年真题

智能刷题助手

智能匹配练习

希望你的学习变得更简单