Data collection gathers information from sources. Surveys, sensors, web logs, transaction records, social media, and public databases all feed the process. The method shapes the data. A survey with leading questions produces biased answers. A sensor with calibration drift produces inaccurate readings. A web scraper that ignores rate limits gets blocked. Collection is not neutral. Every choice affects what you learn.

The first question is what you need and why. Collecting everything is tempting and usually wasteful. Data storage costs money. Processing costs money. Cleaning costs money. Data that sits unused is a liability, not an asset. The second question is how to collect it ethically and legally. Consent matters. Privacy laws like GDPR require a lawful basis for collecting personal data. Terms of service restrict scraping on many platforms. The third question is quality. Is the source reliable? Is the sample representative? Are there gaps? Collection without quality controls produces datasets that fail later. The best time to catch a problem is before the data enters the pipeline, not after someone builds a model on it.

Collection methods

• Surveys — structured questions, prone to response bias
• Sensors — automated, continuous, calibration-sensitive
• Web scraping — extracts public data, subject to legal limits
• Transaction logs — captures user behavior automatically
• Public records — government and open data sources

Data collection is a design decision. What you collect determines what you can learn. What you miss determines what you will never know.