Should I build a Stock Price Predictor, or is that considered a "red flag" project by recruiters?

All answers to this question.

You are 100% correct—avoid the stock market predictor. It’s the "new" Titanic dataset and usually suggests the candidate doesn't understand the Efficient Market Hypothesis. Instead, I highly recommend a "Predictive Maintenance" project. Use a dataset from NASA (like the turbofan degradation set) to predict when a piece of machinery will fail. This uses real sensor data, requires sophisticated feature engineering (like rolling averages and Fourier transforms), and has a clear business value that any industrial or tech company will appreciate.

   Answered 2025-02-22 by Cynthia Higgins

Cynthia, for predictive maintenance, would you recommend using traditional stats models like ARIMA, or should I go straight to deep learning with something like a TCN (Temporal Convolutional Network)?

   Answered 2025-02-24 by George Hall

• George, always start with the baseline! Use a simple Random Forest or XGBoost first. In an interview, if you can show that your fancy TCN only improved accuracy by 1% over a simple XGBoost model, it shows you're a pragmatic engineer who cares about compute costs and model simplicity. That "Baseline vs. Complex" comparison is a great story to tell in a technical interview. It shows you don't just throw deep learning at every problem.

     Commented 2025-02-25 by Cynthia Higgins

"Churn Prediction" is another great one. Every SaaS company on earth needs to know which customers are about to leave. It’s a very hireable project.

   Answered 2025-02-27 by Donna Hayes

• Spot on, Donna. Churn models are the "bread and butter" of many data science teams. If you can build one and explain the "Feature Importance," you're proving you can deliver immediate ROI.

     Commented 2025-02-28 by Paul Rivera