0
$\begingroup$

I am new for data science subject and I am trying to learn on my own. Please bear with me if this is a stupid question. I have a dataset that has no classes. The data set comprises people's activity on Office 365 and my goal is to predict whether the person is experiencing pressure or not using machine learning algorithms, but the dataset does not have classes as you all would expect. Can you please help me how I can solve this problem? Is there any possible way to create classes prior to applying ML models on the data?

Thanks in advance.

$\endgroup$
  • $\begingroup$ Hi, welcome to the forum. It is unclear how you target $y$ looks like. Without this information you cannot expect a proper answer. Please share an example of your data or describe it in more detail $\endgroup$ – Peter Dec 13 '19 at 20:32
2
$\begingroup$

In absence of class labels, your best choices are either an unsupervised machine learning approach (e.g., k-means clustering), or leverage descriptive statistics to see whether you can find any pattern in the data have. For example, looking at the distribution of your features may reveal outliers that could be associated with your outcome of interest.

| improve this answer | |
$\endgroup$
2
$\begingroup$

The data set comprises people's activity on Office 365 and my goal is to predict whether the person is experiencing pressure or not using machine learning algorithms

Very often in data science the problem or the data (or both) is not well defined from the start. Thus the first task is often to design a properly defined ML problem out of the vague initial problem. This is a really important step if one wants to achieve anything useful: ML is not magical, it's a tool which needs to be used in the right way in order to provide interesting results.

According to your description, the initial problem seems ill-defined: for a goal such as predicting whether somebody experiences pressure, there's no clear way to represent "pressure experienced" other than asking the user how much pressure they feel. If your data doesn't contain any such information from the user and you can't obtain it in any other way, then it's likely that the problem cannot be solved at all. As a rule of thumb, if a human cannot find indications in the data which helps answering the question, it's likely that a ML system cannot either.

So your first task is to determine if there's anything in the data that can represent "pressure experienced by the user". For example, maybe a stressed user tends to click everywhere randomly? Then maybe you could try to find in your data if there is a particular pattern or "clicking frequency" and calculate a new column which represents the "stress level" based on this. But there is a high risk that such an assumption is not very reliable, and this would influence your results.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.