Machine Learning Project Description
Using system thinking principles to guide your machine learning experiments allows you to maintain a deeper understanding of the elements that influence intelligent systems and will aid in avoiding the common pitfalls of machine learning projects.
In this series, I will demonstrate purposeful machine learning using a sample time-use dataset. Time-use datasets include snapshots or summaries of how a person spends or allocates their time. We will explore how data can uncover patterns of behavior using a real-world dataset containing both unstructured and structured data. To begin, we will start with a training set of approximately 120K work day summaries (reflects approx. 1000 unique people). A link explaining the process and results of each experiment will be posted in the Experiments table below.
Meanwhile, if you have any questions or input, feel free to post them in the comments area.
- Time period data – The time period in which the activity was performed
- Activity data – activities performed (includes unstructured and structured data)
- Quantitative values – the total time spent on activities during the associated time period
|Phase||Question to be Answered||Learning Type||Training Dataset Size||Results|
|I.||How are activities organized?||Unsupervised, (Clustering)||120K||In progress|
|II.||What kind of person is this? (ex: Administrator, Teacher, Service Provider, Independent Contractor or Other)||Supervised (Classification)||120K||In progress|
|III.||How much of this person’s time will be spent performing administrative duties?||Supervised (Regression)||120K||In progress|
|IV.||Is this an abnormal work day?||Unsupervised (Anomaly Detection)||120K||In progress|