Assuming you are collecting data about traffic accidents in Melbourne in order to develop a predictive model. Would it be better to collect “more data” (e.g. the locations of accidents over many years) or “more types of data” (e.g. the types of vehicles involved, the weather conditions, etc.)? Give a brief justification.[Fill in the blank]简答题
登录即可查看完整答案
我们收录了全球超50000道真实原题与详细解析,现在登录,立即获得答案。

Question

Assuming you are collecting data about traffic accidents in Melbourne in order to develop a predictive model. Would it be better to collect “more data” (e.g. the locations of accidents over many years) or “more types of data” (e.g. the types of vehicles involved, the weather conditions, etc.)? Give a brief justification.[Fill in the blank]简答题

登录即可查看完整答案

我们收录了全球超50000道真实原题与详细解析,现在登录,立即获得答案。

BlackTom AI · Accepted Answer

When collecting data for a predictive model in traffic accidents, there are two broad strategies: increasing the quantity of data (collecting data over many years and across locations) and expanding the variety of data (adding features like vehicle types, weather, road conditions, etc.).
First, consider datasets with more data points over time. More observations can help the model learn patterns across seasons, different locations, and rare events, reducing sampling error. However, if the data points share the same features (no new informational attributes), the model's ability to relate outcomes to inputs may not improve as much as hoped.
Next, adding more types of data increases the dimensionality of the feature space. Each new feature can provide a potential predictor of accidents (e.g., weather conditions, vehicle types, traffic density). This can dramatically improve the model's ability to explain variation and generalize, especially if the existing features are insufficient to capture the underlying drivers of accidents.
That said, there is a caveat: more features can introduce noise and the risk of overfitting, so the data quality and feature engineering matter. In practice, a balance is ideal, but when choosing a strategy in isolation, enriching the feature set with informative, actionable variables generally yields more predictive power per unit of data than merely collecting more observations with the same features.
Additionally, some types of data may unlock interactions and nonlinear effects (e.g., weather conditions interacting with time of day or road type) that raw counts over years cannot reveal without richer attributes.
Thus, the justification for preferring more data types hinges on the idea that models benefit from more informative signals, provided data quality is maintained and appropriate preprocessing is applied.