Boost Sensor Forecast Consistency
Budget: $30 – $250 USD
I’m working with multivariate sensor time-series data and have a baseline model in place. While the average error is acceptable, its performance drifts from day to day. I need a revised approach that steadies the forecasts so trends remain reliable over long stretches.
Scope
• Analyse the existing dataset and baseline metrics I’ll share (CSV, roughly 2 M rows).
• Design or tune a model—whether that’s a smarter feature-engineering pipeline, a hybrid deep-learning architecture (e.g., LSTM with attention), or a classical ensemble—that specifically minimises variance in accuracy across consecutive time windows.
• Implement the solution in Python using common libraries (pandas, NumPy, scikit-learn, TensorFlow/Keras or PyTorch).
• Provide clear, reproducible code plus a short markdown report showing:
– comparison of rolling MAE/RMSE to the baseline,
– any robustness checks (cross-validation, walk-forward tests),
– recommendations for maintaining consistency as new data arrives.
Deliverables
1. Well-commented scripts or Jupyter notebook.
2. Requirements.txt or environment.yml for dependencies.
3. Markdown/PDF report summarising findings and next steps.
I’m open to iterative feedback during the job and will supply sample data immediately.
Scope
• Analyse the existing dataset and baseline metrics I’ll share (CSV, roughly 2 M rows).
• Design or tune a model—whether that’s a smarter feature-engineering pipeline, a hybrid deep-learning architecture (e.g., LSTM with attention), or a classical ensemble—that specifically minimises variance in accuracy across consecutive time windows.
• Implement the solution in Python using common libraries (pandas, NumPy, scikit-learn, TensorFlow/Keras or PyTorch).
• Provide clear, reproducible code plus a short markdown report showing:
– comparison of rolling MAE/RMSE to the baseline,
– any robustness checks (cross-validation, walk-forward tests),
– recommendations for maintaining consistency as new data arrives.
Deliverables
1. Well-commented scripts or Jupyter notebook.
2. Requirements.txt or environment.yml for dependencies.
3. Markdown/PDF report summarising findings and next steps.
I’m open to iterative feedback during the job and will supply sample data immediately.