Hi folks! In the previous video, we generated some synthetic banking data and began to explore its representativeness compared to the original production data. We found that the Privacy synthetic data tells the same story as the original data. In this video, we will take a look at a second use case using the same synthetic banking data. We will upload it to ChatGPT code interpreter for exploration and analysis.
ChatGPT Plus offers a beta version of code interpreter that allows file uploads. I have saved my bank marketing synthetic data and will use it to perform various tasks. First, I will ask ChatGPT to give me a profile of the data set, create basic graphs and visualizations, and identify the three key variables impacting the classification goal.
Once the analysis is done, I will use code interpreter to validate my next synchronization on the platform. I want to upsample the duration of calls in my data set to understand the average length of successful calls. ChatGPT tells me it’s 540 seconds. I will upload the modified data set to the platform and rebalance the ‘duration high’ column, which indicates calls that are 540 seconds or longer.
The most AI platform allows me to upsample the instances where call durations are 540 seconds or above. I will upsample it to 50 instances and launch the job. After the job is completed, I will review the QA report to see how closely the synthetic data matches the original data. In the ‘duration high’ column, I now have approximately 50% instances of high duration calls in the synthetic data compared to roughly 11% in the original data.
This increase in high duration calls has positively impacted the number of successes in my direct marketing campaign. It’s fascinating to see how synthetic data can be used to tell a different story from the original data. This demonstrates the power of code interpreter and the benefits of using synthetic data for data democratization.
Thank you for watching!
