Upload anything → Get anomaly alerts & forecasts

Production-ready REST API with:

• 📝 Log Anomaly Detection (TF-IDF + DBSCAN for text)
• 🔍 Outlier Detection (IQR, Z-score, Isolation Forest)
• 📈 Time-Series Forecasting (Linear, Exponential, Prophet)
• 🧠 Unified Model (One model, all data, continuous learning)
• 📊 Smart Upload (Auto-detects CSV, Excel, JSON, Logs)

55,000+ rows tested | 6 datasets | 1 unified model

# Install & generate data
make install
make data

# Analyze a dataset (NO CURL NEEDED!)
python app.py analyze data/sensor_data.csv temperature

# Train unified model
python app.py train data/sales_data.csv sales

# Check status
python app.py status

# Start server
python app.py

# Or use Makefile shortcuts
make analyze-sensors  # Analyze 20K sensor readings
make train            # Train on all 55K samples
make help             # See all commands

✨ Best-in-class lightweight algorithms:

• Outlier Detection: IQR, Z-score, Isolation Forest
• Linear Regression: Ridge regression with automatic feature scaling
• Time-Series Forecasting: Exponential smoothing & linear trend
• Model Training & Persistence: Train once, predict many times

🚀 Production-ready:

• Thread-safe data & model storage
• Rate limiting & security headers
• Prometheus metrics
• Health checks & monitoring
• Docker support

pip install -r requirements.txt

python app.py

Server starts at http://localhost:5000

python test.py

Use the ready-to-go datasets in sample_data/ to exercise every endpoint without having to craft payloads from scratch:

1. Upload the CSV/JSON data

   python - <<'PY'
   import pandas as pd, requests
   df = pd.read_csv('sample_data/sales_data.csv')
   payload = {"name": "sales_sample", "data": df.to_dict('records')}
   res = requests.post('http://localhost:5000/upload_data', json=payload, timeout=10)
   res.raise_for_status()
   print(res.json())
   PY

2. Detect outliers:

   curl -s http://localhost:5000/detect_outliers \
     -H 'Content-Type: application/json' \
     -d '{"dataset": "sales_sample", "method": "iforest", "columns": ["sales", "marketing_spend"]}' | jq .statistics

3. Run regression:

   curl -s http://localhost:5000/regression \
     -H 'Content-Type: application/json' \
     -d '{"dataset": "sales_sample", "target": "sales", "features": ["marketing_spend", "temperature"], "model_name": "sales_model"}' | jq .metrics

4. Forecast with the energy dataset:

   python - <<'PY'
   import pandas as pd, requests
   df = pd.read_csv('sample_data/energy_usage.csv')
   payload = {"name": "energy_sample", "data": df.to_dict('records')}
   requests.post('http://localhost:5000/upload_data', json=payload, timeout=10).raise_for_status()
   res = requests.post('http://localhost:5000/forecast', json={
       "dataset": "energy_sample",
       "value_column": "energy_usage",
       "method": "linear",
       "horizon": 5
   }, timeout=10)
   print(res.json()["forecast"])
   PY

5. Run the automated suite once the data is loaded:

   pytest tests/test_analytics.py -v

POST /upload_data
{"name": "sales", "data": [{"col1": 10, "col2": 20}]}

POST /detect_outliers
{"dataset": "sales", "method": "iqr"}
# methods: iqr, zscore, iforest

POST /train_model
{"dataset": "sales", "target": "col1", "model_name": "predictor"}

POST /predict
{"model_name": "predictor", "data": [{"col2": 25}]}

POST /forecast
{"dataset": "sales", "value_column": "col1", "method": "linear", "horizon": 10}
# methods: linear, exponential

POST /regression
{"dataset": "sales", "target": "col1", "features": ["col2"]}

import requests

# Upload data
requests.post('http://localhost:5000/upload_data', json={
    "name": "sales",
    "data": [
        {"date": "2024-01-01", "sales": 100, "marketing": 20},
        {"date": "2024-01-02", "sales": 120, "marketing": 25}
    ]
})

# Detect outliers
result = requests.post('http://localhost:5000/detect_outliers', json={
    "dataset": "sales",
    "method": "iforest"
})
print(f"Found {result.json()['n_outliers']} outliers")

# Train model
requests.post('http://localhost:5000/train_model', json={
    "dataset": "sales",
    "target": "sales",
    "model_name": "predictor",
    "features": ["marketing"]
})

# Predict
predictions = requests.post('http://localhost:5000/predict', json={
    "model_name": "predictor",
    "data": [{"marketing": 30}]
})
print(predictions.json()['predictions'])

Create .env file (see .env.example):

SECRET_KEY=your-secret-key
RATE_LIMIT_PER_MINUTE=60
MAX_BATCH_SIZE=10000
MODEL_DIR=models

docker-compose up -d

IQR: Fast, no training. Flags values outside Q1-1.5×IQR to Q3+1.5×IQR
Z-Score: Assumes normal distribution. Flags |z| > 3
Isolation Forest: Best for high-dimensional, non-linear patterns
Ridge Regression: L2 regularization, automatic scaling
Exponential Smoothing: Fast, good for stationary data
Linear Trend: Simple, interpretable forecasting

• 100K rows in memory
• Sub-second responses
• Isolation Forest: ~100ms for 10K rows
• Regression: ~50ms for 10K rows

MIT