Behind the Screen

Predict the Burning Behind the Screen

Comment

Inspiration

As a Harvard Gazette article describes, long after vaccines have tamed COVID-19’s physical impacts, its mental health effects will linger. Social distancing and work-from-home guidelines have only exacerbated issues of mental health amid rising feelings of distress, likely stemming from people’s limited social interactions, tensions among families in lockdown together, and fear of illness. Indeed, as Nature reports, more than 42% of people surveyed by the US Census Bureau in December reported symptoms of anxiety or depression in December, an increase from 11% the previous year. Therefore, now more than ever before, it’s important that we have a means of identifying symptoms of mental health issues.

Specifically, employee burnout, a state where one loses all motivation or incentive, is on the rise. This can lead to feelings of stress and even depression. Burnout usually results as a product of high stress or fatigue and is often accompanied by feelings of powerlessness and hopelessness.

Furthermore, when workers feel physically or emotionally exhausted, they're not able to participate fully in the workplace or even at home. A 2018 Gallup poll found astounding consequences for employee burnout. Employees who say they very often or always experience burnout at work are: 13% less confident in their performance Half as likely to discuss how to approach performance goals with their manager 2.6 times as likely to leave their current employer 63% more likely to take a sick day 23% more likely to visit the emergency room

Our project combats this issue by identifying employee burnout so that such individuals can get the help they need.

What it does

With scikit-learn’s DecisionTreeRegressor, our project predicts the employee burnout on a scale from 0-1 given three features: seniority in their respective institution on a scale from 0-5, the amount of resources allocated to the employee to work (i.e. number of working hours) on a scale from 1-10, and a self-reported mental fatigue score on a scale from 0-10.

Special thanks to HackerEarth, “Are Your Employees Burning Out?” Kaggle, 2020, doi: 10.34740/KAGGLE/DS/949779. for the dataset.

How we built it

After gathering our data from HackerEarth in a pandas Dataframe, we looked specifically at the Designation (the individual’s seniority within their institution), Resource Allocation, and Mental Fatigue Score parameters in train.csv. These features constituted the "x" part of our (x,y) data. At first, we used the categorical parameters Date of Joining, Gender, Company Type, and WFH Setup Available as well, but removed these features from our dataset once our results suggested they provided negligible information gain. For the "y" part of our (x,y) training data, we extracted the Burn Rate parameter from train.csv.

Next, we trained a DecisionTreeRegressor on 70% of our total data (the remaining 30% served as our testing set) with scikit-learn. Decision trees are a supervised learning algorithm, meaning that given knowledge of an individual’s seniority, the amount of resources allocated to that individual, and the self-reported mental fatigue score, we can predict the burn rate. Furthermore, we tried averaging the accuracy of multiple decision trees (bagging) and constructing a series of trees that build upon the previous tree’s weaknesses (boosting), ensemble methods known as RandomForestRegressor and GradientBoostingRegressor respectively, as well, however, these improved accuracy by only about 1%.

Challenges we ran into

Although our team is well-versed in Python, we had not previously applied these skills in a real-world setting with artificial intelligence libraries like scikit-learn. Encoding categorical data, for example, was something we were unfamiliar with and had to research. Additionally, while we originally tried to host our website on Glitch, this made it difficult to connect our Python-based decision trees to our JavaScript-based Express application (which is why we switched to hosting the application on our local computers).

Accomplishments that we're proud of

We applied the skills we learned in classroom settings in a real-world context. Not only did we actively search for and find reliable data, we also creatively analyzed that data to create social impact. By taking initiative for our own learning, we developed critical thinking and problem solving skills that will help us continue to create technology for social good in the future.

What we learned

We learned how important it is to be conscious of not only our own mental health, but also the mental health of those around us. By ignoring symptoms of burnout, we only hurt ourselves and the productivity of our work. It is essential that we bring awareness to the mental health issues that exist in the workplace and beyond.

By designing and developing this website, we also refined our programming skills and learned new techniques. In particular, we learned how to use DecisionTreeRegressors on real-world data and create a meaningful website for social good.

What's next for Behind the Screen

Behind the Screen can be expanded to a mobile application that would alert users for when they may have an increased risk of burnout. The mobile application would also inform users of possible steps they could take to alleviate and avoid that burnout. Our current website provides a foundation for such extensions that would prove to be valuable to businesses and their employees.

Built With

Share this project:

Updates