ML Time-Series Signal Processing

I need a machine-learning expert who can take raw time-series signals and turn them into accurate, production-ready models. The entire effort revolves around signal processing—filtering, denoising, feature extraction—and then building, training, and validating models that make reliable predictions from those signals. The data you will handle arrives as multichannel time-series streams. You will decide on the best preprocessing pipeline (e.g., wavelet or FFT filtering, adaptive smoothing, normalization), engineer features, and then implement and compare models such as CNNs, transformers, or traditional algorithms if they outperform deep nets for this context. Python is my preferred stack; NumPy, SciPy, scikit-learn, PyTorch or TensorFlow should feel second nature to you. Robust documentation and clear, reproducible notebooks/scripts are mandatory so the pipeline can be audited and extended later. To be considered, include a detailed project proposal that outlines: • the end-to-end workflow you intend to follow, • the specific algorithms or architectures you would start with and why, • the evaluation strategy (cross-validation approach and metrics), • an estimated timeline broken into milestones. Deliverables I expect: 1. Clean, well-commented code or notebooks covering preprocessing through model deployment. 2. A concise technical report summarizing methods, results, and next-step recommendations. 3. A short hand-off session (recorded video or live call) walking me through the pipeline and highlighting retraining steps. I will review proposals on their clarity, feasibility, and how well they anticipate common pitfalls in time-series ML such as non-stationarity, class imbalance, and overfitting.

Project ID: 40517496