3.md

• Daily note taking moved from Notes on SE Searches and Meeting Notes in Google Drive to this calandar.

• Decisions made :

  • Stopped watching lectures since it doesn't contribute to narrow down thes search for goal making process
  • Selecting a specific senario is important. since nature of data dictates techquniqes used
  • Found this slide set
  • Fields to select from
    • Banking - too much competition
    • Finance - lot of statistical techquniques
    • Socila Media - Intresting data set

• Skimmed Anomaly detection on time series to get more direction.

  • Problem setting for time series anomaly detection

    • Detecting contextual anomilies
    • Detecting anomalous subsequencewith respect to a given long sequence (time series)
    • If the anomalous subsequence is of unit length, this problem is equivalent to finding contextual anomalies in the time series

  • Chalenges

    • exact length of sumsequence is not known

  • Types of time series data

    • Periodic - Synchronous(multiple values are synchronised eg: temprature & pressue messurements are in sync) time series
    • Aperiodic - Synchronous
    • Periodic - Asynchronous
    • Aperiodic - Asynchronous

  • AD techniques can be classified by process(Procedural dimension) and the way data transformed prior to AD (Transformation dimension)

  • New leads:

    • To be clarified by on next work session