A time series is a repeated measurement recorded at different, ideally uniform, time intervals. Its rows usually pair a date or time with one or more measurement values; time makes each observation distinct. GDP, machine output, customer-record creation, satisfaction scores, and traffic counts can all become time series when examined over time. section 8.0 section 8.1
Table 8.1 An example of a time series dataset
|
Date
|
Satisfaction score
|
|---|---|
|
2023-11-01 11:03:55
|
4
|
|
2023-11-01 11:17:02
|
5
|
|
2023-11-01 13:41:11
|
3
|
|
2023-11-01 14:06:43
|
4
|
A timestamp and satisfaction score are enough to support questions about activity levels, trends, seasonal patterns, anomalies, and differences between locations. section 8.1.1
Treat time series work as both data preparation and analytical judgment. First establish what one row represents, whether observations form one series or many, their time granularity, coverage, completeness, duplicate records, and gaps. Then choose a resolution that fits the decision: hourly values may be noisy, while daily or monthly aggregation can hide meaningful patterns. section 8.1.2 section 8.4.1 section 8.5
Common questions for a time series include:
- How does the measurement change over time, across locations, or at daily, weekly, and monthly levels? section 8.1.1
- Are there trends, seasonal patterns, peak periods, unexpected spikes, or anomalous observations worth investigating? section 8.1.2 section 9.3.1
- Do past values inform future values, and is there sufficient gap-free history for a useful forecast? section 8.1.2 section 9.4
- Can external factors explain unusual values or improve a forecast? section 8.1.2
Selecting cycling infrastructure candidates
Traffic records contain many series rather than one: each count point can be analyzed separately, or measurements can be aggregated by region, authority, or period. For cycling decisions, define whether “upgrade” means high existing traffic, fastest growth, or forecast growth; retain raw counts when area-level summaries are too coarse. Check coverage and gaps before comparing locations or forecasting them. section 8.2.2 section 8.3 section 8.4.1
Forecasting is not the starting point or the only outcome. Decide how forecasts will be used, including required frequency, horizon, granularity, acceptable accuracy, business value, and the return on further improvement. Start with fundamentals before relying on more complex forecasting tools, and evaluate a model to determine whether the series is forecastable. section 8.1.2 section 9.3 section 9.4
