Job Openings Prediction during and post COVID-19

Inspiration

Statistics Canada reported that since the pandemic the jobless rate held steady at 8.9% representing 1.8 million Canadians who are officially classified as unemployed, whereas the jobless rate prior to the pandemic (February) was 5.6% (CBC, 2020). As active job seekers, we are interested in exploring changes in the job openings in Canada since the pandemic to understand the industry trend and help navigating the opportunities and challenges in the job search.

CBC. (2020, November 6). Job market recovery from COVID-19 slows in October, with only 84,000 new jobs Social Sharing. CBC News. https://www.cbc.ca/news/business/jobs-canada-october-1.5792149

What it does

The regression model is fed in weekly job posting data taken from Burning Glass (https://www.burning-glass.com/research/open-data-job-postings/?fbclid=IwAR38PnjSaZYtCB06FmR8-ffBuwaaFyWuwEWczewcLot7KHUkn_-0Wg7GsRs) and is expected to predict a weekly job opening number by industry during and post the pandemic.

How we built it

There are two major activities involved in this project: data cleaning and model building. The dataset has been subsetted to Canada only and top 10 industries are selected based on the number of job openings and our interests. To check if the date of job listings affects the number, start date and end date have been transformed into separate year, month and date attributes. After fitting and transforming the train set, the regression model is built using the tensorflow package. Mean squared error is chosen as the loss metric for the model.

Challenges we ran into

Throughout this hackathon, there were multiple challenges that our group faced. First being, prior to even gathering our dataset we had a sudden occurrence of 2 members dropping out. This significantly impacted us as the 2 remaining members were complete beginners with the only guide being from the workshops. The next challenge we faced was deciding on a dataset and what we wanted our ML model to learn. As the topic for this hackathon was fairly broad, it took looking at multiple datasets to come up with the idea of predicting how jobs were affected due to COVID-19. Lastly, cleaning the dataset and building the model was a serious challenge as it was our first time using pandas and tensorflow. Our model also faced an issue of only obtaining a 4% accuracy rate which by then we realized it was too late to readjust our dataset making for an extremely complicated task.

Accomplishments that we're proud of

Even though our model had an extremely low accuracy rate, as complete beginners who did not have any hands on experience with ML prior to the workshops we were able to obtain our dataset, clean it, and run a model on it. That itself was a significant accomplishment and is something we plan on working with to improve over the next few months.

What we learned

From a general view, this hackathon helped us learn more pandas, tensorflow, seaborn and various functions that it offers.Our group also learned that having a dataset with sufficient predictors is a must which we only ended up finding out about after running our model and obtaining a 4% accuracy rate.

What's next for Job Openings Prediction during and post COVID-19

To further improve our model, some next steps for us would be to rework the dataset and find more predictors such as covid confirmed cases, geolocation of the job openings, economic factors and lockdown measures, etc. In addition we also plan on testing various prediction models and finding one particular model that would work more closely to what we actually want to predict.

Built With

• colab
• python