Chronoamperometry Pulse Classification

Classifies 7 object types (I–VII) from chronoamperometry (CA) CH1 pulse signals. Each class has 100 repeated pulses; the pipeline segments them, extracts pulse-shape features, trains an SVM, and produces a t-SNE cluster diagram and a confusion matrix.

Layout

.
├── main.py                  # entry point
├── requirements.txt
├── src/
│   ├── config.py            # paths, class→file→sampling-rate mapping
│   ├── data_loader.py       # UTF-16LE CSV reader, CH1 only
│   ├── segmentation.py      # find_peaks-based 100-pulse cutter
│   ├── features.py          # 11 per-pulse features
│   ├── classification.py    # StandardScaler + SVM(RBF), 10-fold CV
│   ├── visualization.py     # t-SNE + confusion-matrix figures
│   └── pipeline.py          # orchestrates the full run
├── outputs/                 # generated
│   ├── features.csv
│   ├── classification_report.txt
│   ├── tsne_cluster.png
│   ├── confusion_matrix.png
│   └── debug/segmentation_<class>.png  # peak overlays for sanity-check
└── *.csv                    # 7 input recordings

Run

pip install -r requirements.txt
python main.py

Notes

• CH1 only (columns 0 and 1); UTF-16LE with a 6-line header.
• Sampling: 0.01 s for 25_*.csv, 0.1 s for high_*.csv / higher_*.csv (see src/config.py).
• Segmentation cuts at midpoints between the 100 most prominent peaks, so every sample belongs to exactly one pulse and every window owns one peak.
• The confusion matrix is built from cross_val_predict (held-out folds), never from training-set predictions.
• All randomness pinned with random_state=42.