Add Dockerfile and README

Dockerfile

@@ -0,0 +1,10 @@
+FROM python:3.11-slim
+RUN useradd -m -u 1000 user
+USER user
+ENV PATH="/home/user/.local/bin:"
+WORKDIR /app
+COPY --chown=user requirements.txt requirements.txt
+RUN pip install --no-cache-dir --upgrade -r requirements.txt
+COPY --chown=user . /app
+EXPOSE 7860
+CMD ["gunicorn", "--bind", "0.0.0.0:7860", "--workers", "2", "--timeout", "120", "app:app"]

README.md

@@ -1,313 +1,13 @@
 ---
-title: Fault Detection For Industry
-emoji: 🌍
-colorFrom: indigo
-colorTo: gray
+title: FaultSense
+emoji: ??
+colorFrom: green
+colorTo: blue
 sdk: docker
+app_file: app.py
+app_port: 7860
 pinned: false
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
-# FaultSense — Industrial Equipment Fault Predictor
-
-> Real-time binary fault detection for industrial equipment using LightGBM, served via a Flask web application.
-
-![Python](https://img.shields.io/badge/Python-3.9%2B-blue)
-![LightGBM](https://img.shields.io/badge/Model-LightGBM-brightgreen)
-![Flask](https://img.shields.io/badge/API-Flask-lightgrey)
-![License](https://img.shields.io/badge/License-MIT-yellow)
-
----
-
-## Overview
-
-FaultSense takes live sensor readings from industrial equipment — temperature, pressure, vibration, and humidity — and predicts in real time whether the equipment is **healthy** or **faulty**. It includes a full ML pipeline from synthetic data generation through hyperparameter search to a production-ready web interface.
-
-**Equipment types supported:** Pump · Compressor · Motor · Valve · Sensor
-
----
-
-## Screenshots
-
-The web UI lets you drag sensor sliders and get an instant fault prediction with probability score and confidence level.
-
-> Run the app (see below) and open `http://localhost:5000`
-
----
-
-## Project Structure
-
-```
-FaultSense/
-│
-├── app.py                          # Flask web app (main entry point)
-├── app2.py                         # Alternative app variant
-│
-├── data_synthesier.py              # Synthetic dataset generator
-├── dataset.py                      # Dataset structuring utilities
-├── distribution_function.py        # Sensor feature distribution modelling
-├── data_analyze.py                 # Exploratory data analysis
-├── data.ipynb                      # EDA notebook
-│
-├── main.py → main8.py              # Iterative experiment scripts
-├── main9_by_claude.py              # Claude-assisted experiment
-├── main10_claude_combnation.py     # Dense hyperparameter grid search (~13,650 runs)
-├── main11.py                       # Final experiment iteration
-│
-├── synthetic_nim_parallel_10000.csv  # Primary training dataset (10,000 samples)
-├── RANDOM_FOREST.csv               # Random Forest baseline results
-├── faultsense_model.joblib         # Serialised trained pipeline
-│
-├── results/                        # Experiment results (CSV / XLSX)
-├── plots/                          # Saved diagnostic plots
-├── analysis/                       # Additional analysis outputs
-├── industrial-equipment-monitoring-dataset/  # Raw dataset folder
-└── synthetics3/                    # Additional synthetic data variants
-```
-
----
-
-## Features
-
-- **Binary fault classification** — predicts `FAULTY` or `HEALTHY` with probability score
-- **Confidence levels** — HIGH / MEDIUM / LOW based on prediction probability
-- **Live web UI** — interactive sliders for all sensor inputs, dark-mode interface
-- **Prediction history** — last 20 predictions shown in-session
-- **Model info panel** — displays test AUC, F1, accuracy, precision, and recall live in the UI
-- **REST API** — `/predict` endpoint accepts JSON for programmatic use
-- **Auto train or load** — automatically retrains if no saved model is found
-
----
-
-## Quickstart
-
-### 1. Clone the repository
-
-```bash
-git clone https://github.com/techavenger123/Trial_AI_ProjDATASET.git
-cd Trial_AI_ProjDATASET
-```
-
-### 2. Install dependencies
-
-```bash
-pip install -r requirements.txt
-```
-
-### 3. Run the app
-
-```bash
-python app.py
-```
-
-On first run, if `faultsense_model.joblib` is not present, the model will train automatically using `synthetic_nim_parallel_10000.csv`. This takes under a minute.
-
-### 4. Open in browser
-
-```
-http://localhost:5000
-```
-
----
-
-## Requirements
-
-```
-flask
-lightgbm
-scikit-learn
-pandas
-numpy
-joblib
-matplotlib
-tqdm
-openpyxl
-```
-
-Install all at once:
-
-```bash
-pip install flask lightgbm scikit-learn pandas numpy joblib matplotlib tqdm openpyxl
-```
-
-Python 3.9 or higher is recommended.
-
----
-
-## Dataset
-
-The primary dataset (`synthetic_nim_parallel_10000.csv`) contains **10,000 synthetic sensor readings** generated in parallel to simulate realistic industrial conditions.
-
-| Feature | Type | Range | Description |
-|---|---|---|---|
-| `equipment` | Categorical | pump, compressor, motor, valve, sensor | Equipment type |
-| `temperature` | Float | –20 to 120 °C | Operating temperature |
-| `pressure` | Float | 0 to 20 bar | Internal pressure |
-| `vibration` | Float | 0 to 50 mm/s | Vibration level |
-| `humidity` | Float | 0 to 100 % | Ambient humidity |
-| `location` | Categorical | — | Installation location (dropped at training) |
-| `faulty` | Binary | 0 / 1 | **Target** — 0 = healthy, 1 = faulty |
-
-Class imbalance is handled via `class_weight="balanced"` in the LightGBM classifier.
-
----
-
-## Model
-
-FaultSense uses a **scikit-learn Pipeline** combining preprocessing and a LightGBM classifier.
-
-### Architecture
-
-```
-Input features
-    │
-    ├── equipment (categorical) ──► OneHotEncoder
-    └── temperature, pressure,
-        vibration, humidity (numeric) ──► passthrough
-                        │
-                        ▼
-                LGBMClassifier
-                        │
-                        ▼
-              Fault probability [0–1]
-                        │
-                  threshold = 0.5
-                        │
-              FAULTY (1) / HEALTHY (0)
-```
-
-### Best configuration
-
-| Parameter | Value |
-|---|---|
-| Learning rate | 0.05 |
-| n_estimators | 165 |
-| max_depth | 8 |
-| num_leaves | 50 |
-| subsample | 0.8 |
-| colsample_bytree | 0.8 |
-| Train / Val / Test split | 90% / 5% / 5% |
-| Prediction threshold | 0.5 |
-
-This configuration was selected from a dense grid search of **~13,650 combinations** across 35 learning rates, 78 estimator counts, and 5 train/val/test split ratios (see `main10_claude_combnation.py`).
-
----
-
-## API Reference
-
-### `POST /predict`
-
-Predict fault status from sensor readings.
-
-**Request body (JSON)**
-
-```json
-{
-  "equipment": "pump",
-  "temperature": 75.5,
-  "pressure": 12.3,
-  "vibration": 18.0,
-  "humidity": 65
-}
-```
-
-**Response**
-
-```json
-{
-  "prediction": 1,
-  "probability": 0.8732,
-  "confidence": "HIGH",
-  "threshold": 0.5,
-  "label": "FAULTY"
-}
-```
-
-### `GET /model_info`
-
-Returns the current model configuration and test-set performance metrics.
-
-**Response**
-
-```json
-{
-  "config": {
-    "learning_rate": 0.05,
-    "n_estimators": 165,
-    "train_ratio": 0.9,
-    "val_ratio": 0.05,
-    "test_ratio": 0.05
-  },
-  "test_metrics": {
-    "test_auc": 0.97,
-    "test_accuracy": 0.94,
-    "test_f1": 0.93,
-    "test_precision": 0.91,
-    "test_recall": 0.95,
-    "test_logloss": 0.18
-  }
-}
-```
-
----
-
-## Running the Hyperparameter Search
-
-To reproduce the full grid search (warning: this takes significant time — ~13,650 model fits):
-
-```bash
-python main10_claude_combnation.py
-```
-
-Results are saved to `results/synthetic/dense_results.csv` and `dense_results.xlsx`. Six diagnostic plots are saved to `Synthetic1/synthetic_plot/`:
-
-- Validation metric heatmaps (LR × n_estimators)
-- Metrics vs n_estimators per split ratio
-- Metrics vs learning rate per split ratio
-- Train vs validation curves (best split)
-- Overfitting heatmap (train AUC − val AUC)
-- Top-30 config scatter (val F1 vs val AUC)
-
-The search supports **checkpointing** — if interrupted, it resumes from where it left off.
-
----
-
-## Retrain from Scratch
-
-To force a full retrain (ignoring any saved model):
-
-```bash
-# Delete the saved model, then run the app
-rm faultsense_model.joblib
-python app.py
-```
-
-Or edit `BEST_CONFIG` in `app.py` to change hyperparameters before retraining.
-
----
-
-## Known Limitations
-
-- **Synthetic data only** — the model has not been validated on real industrial sensor readings. Performance may differ on real-world data.
-- **Fixed threshold** — the prediction threshold is set to 0.5. For safety-critical applications, consider tuning this using a precision-recall curve to favour recall (catching more faults at the cost of more false alarms).
-- **No feature explainability** — the app does not currently show which sensor reading drove a given prediction. Adding SHAP values would improve interpretability for maintenance engineers.
-- **No authentication** — the Flask app runs without any access control. Do not expose it publicly without adding authentication.
-- **Single model** — only LightGBM is deployed. Ensemble approaches or periodic retraining on fresh data may improve production reliability.
-
----
-
-## Development History
-
-This project was built iteratively, with experiment scripts versioned as `main.py` through `main11.py`. Scripts `main9_by_claude.py` and `main10_claude_combnation.py` reflect AI-assisted development using Claude.
-
----
-
-## License
-
-MIT License. See `LICENSE` for details.
-
----
-
-## Contributing
-
-Pull requests are welcome. For significant changes, please open an issue first to discuss what you would like to change.
+# FaultSense � Industrial Equipment Fault Predictor
+Real-time binary fault detection using LightGBM and Flask.