Machine Fault Detection is a subfield of predictive maintenance. Data-driven based approaches are becoming popular with the development of deep learning. Researchers and practitioners have been trying to apply deep learning in machine fault detection. However, a large number of labeled data, especially data for machines in failure, are needed to achieve a good performance.
The fault data is usually generated by simulation: manually breaking the machine and then record the data. However, for a machine, there are infinitely many ways for it to run in malfunction. Simulating all possible faults is too costly and time-consuming.
Here comes the problem…
Earnings conference call is a question-answering session between executives and major investors where company comments on its financial results. It is usually held quarterly by most publicly traded corporations. Some researchers have tested that earnings conference call transcripts has some prediction power on stock price movement. However, the stock price is not the only thing to focus on predicting while making use of earnings calls. Predicting credit risk is particularly complicated and difficult. This led to my attempt to build a binary text classifier which identify possible credit risk given the earnings call transcript of the company issuing the bond.
Students refer to Faculty Course Evaluation (FCE) when selecting courses to take.
In my analysis, I attempted to do some data manipulation and correlation analysis using the given data, to find out: the most high-rated professors and courses in Stat&ML so that I can plan a (somewhat) easier semester :) And also what features in FCE are most correlated to the Overall course rate, which can be a suggestion for faculty to improve FCE ratings.
Congratulations to Joel (my 401 professor)!