June 10, 2025
Day 11- Changing Direction
What I Learned
Today we encountered a significant issue with the dataset we had been using for our project. During our analysis, we realized that the dataset consisted entirely of 2D ECG images, which posed a major problem: it’s extremely challenging to extract precise numerical data from image-based ECG recordings. This is because ECG images are visual representations rather than raw numerical time-series data, and any attempt to extract signals from these images can lead to inaccuracies, loss of resolution, and potentially misleading results. This discovery forced us to rethink our data strategy. After careful consideration and discussions with the team, we decided that the most effective solution would be to switch to a new dataset that natively contains numerical ECG data in time-series format. This kind of dataset is much better suited for our planned machine learning models, including 1D CNNs and LSTMs, which require raw sequential data to capture the temporal dynamics of heartbeats accurately. As a result, today was primarily spent researching, evaluating, and selecting a new dataset that aligns with our project goals. We compared several options and considered factors like the availability of raw ECG signals, the comprehensiveness of diagnostic labels, and the potential for improving model accuracy. Ultimately, we settled on a dataset that provides high-quality numerical ECG data, which will allow us to train our models more effectively and potentially achieve better performance in classifying different heart conditions. In summary, while it was a challenging day due to the setback we faced, it was also a productive one because we made a critical decision that will improve the integrity and performance of our entire project going forward.
Blockers
No Blockers
Reflection
Today taught me something important about research—and even about life in general. It reminded me that no matter how far ahead you think you are, there will always be moments when unexpected obstacles force you to change direction. It’s a valuable lesson: to always stay open and flexible, ready to pivot when new information comes to light. I hadn’t really anticipated that we’d need to switch the entire dataset, especially after the progress we’d made. But when we discovered that our dataset was based on 2D ECG images rather than raw numerical ECG signals, we had no choice but to adapt. Sticking with the old dataset would have meant continuing with an approach that wouldn’t yield the accuracy and results we needed. So we made the tough decision to change it, despite the effort and time it will take to realign our project. It’s a good reminder that progress isn’t always a straight line. Sometimes it requires stepping back, re-evaluating, and making a shift that might seem like a setback at first—but ultimately sets you on a better path.