Maswali ya Kujifunza Isiyosimamiwa

Majibu na maelezo huonekana unapojibu maswali.

1. A dataset has no labels, but a team wants to group similar records. Which family of methods fits that goal?
   • Kujifunza bila kusimamiwa
   • Uainishaji unaosimamiwa
   • Target encoding
   • Regression
2. Two customer records differ by one visit and 1,000 dollars of spending. What can raw Euclidean distance do?
   • Automatically treat the features equally
   • Be dominated by the spending difference because feature scales differ
   • Prove that the records belong to different real categories
   • Remove the need to inspect features
3. A two-dimensional projection shows two separated clouds. What limitation should an analyst remember?
   • Every distance in the plot exactly preserves all original distances
   • The projection proves the groups are true categories
   • The projection discards information, so apparent gaps are not unquestionable facts about the original data
   • Visualization removes the need to inspect raw dimensions
4. A clustering result changes after reasonable preprocessing choices. What should the team do?
   • Treat preprocessing as irrelevant
   • Name the first result as a discovered truth
   • Choose whichever plot looks most separated
   • Examine stability and inspect representative and borderline records
5. A cluster is built from purchase frequency rather than product preferences. What follows?
   • The cluster names are fixed by the algorithm
   • The similarity rule and features affect what groupings appear
   • Both analyses must return identical segments
   • Purchase frequency proves customer intent
6. An internal clustering score is high. What does that establish?
   • That the cluster will remain stable in all settings
   • That the group is a true human category
   • A mathematical property under that score, not automatic usefulness for the real purpose
   • That anomalies are harmful
7. An anomaly detector flags an unusual transaction. What can the alert support?
   • Investigation, but not proof of misconduct
   • An automatic finding of fraud
   • A diagnosis of the account owner
   • A guaranteed reason for the unusual value
8. A team scales features and sees different neighbors. Why is scaling relevant?
   • It makes high-dimensional data lossless
   • It supplies target labels
   • It proves a better category system
   • It changes how feature differences contribute to similarity