DataEngBytes 2023 - MEL-T3-01 - Rajath Vanikul

on YouTube ↗23m

" summary

Binary classifier detects electric vehicles from smart‑meter consumption by reducing four weeks of time series into monthly customer‑level snapshots and engineered stats, iterating through edge cases (hot‑water heaters, HVAC temperature effects, weekend holiday gaps) until a deployed model runs in prod. Pipeline uses weather overlays, a feature store to left‑shift expensive transforms (cutting repeated compute after a 92% data‑fetch/engineering bottleneck), plus Champion‑Challenger model registry and automated comparison metrics.

AI-written summary · about

" chapters (12)

1. 0:00Sustainable ML models intro
2. 2:30ML architecture fundamentals
3. 4:55EV detection use case
4. 8:29EV consumption patterns
5. 10:38Model evolution Spider-Man 1-3
6. 13:52Model challenges solutions
7. 14:34Continuous improvement in DS
8. 17:44Ensemble learning techniques
9. 19:07Model registry importance
10. 19:50Left shift feature store
11. 21:36Automated model comparison
12. 22:42Takeaways and conclusion

" related talks

• 26mDataEngBytes 2023 - SYD-T2-03 - Akanksha MalikAkanksha Malik · DataEngBytes Sydney 2023
• 21mDataEngBytes 2023 - SYD-T2-04 - Suneeta MallSuneeta Mall · DataEngBytes Sydney 2023
• 40mDeep Learning and Time Series Forecasting for Smarter EnergyIgor Gotlibovych · PyData London 2019
• 38mElectrifying Big Issues Through AnalyticsMukund Kaushik · Data Con LA 2019
• 21mDataEngBytes 2023 - SYD-T2-02 - Vinny Vijekumaar & Zak SheikhVinny Vijekumaar, Zak Sheikh · DataEngBytes Sydney 2023
• 48mIdentification of Building Thermal Model - Using DataMihael Jaksic · DSC Europe 23
• 20mDataEngBytes 2023 PER-T1-10 - Matthew CarterMatthew Carter · DataEngBytes Perth 2023
• 11mData Science and Smart Systems: Creating the Digital BrainKaushik Das · Strata Conference 2014