DPS-AI-Model

https://dps-ai-model-ishq.onrender.com/predict - run it on postman api

SARIMA-Based Forecasting API

This project is a Flask-based API that provides predictions using time-series forecasting models. It allows users to input a specific year and month and returns the predicted value for that period.

---

Features

• Implements SARIMA, Prophet, and a Hybrid model (combining both SARIMA and Prophet) for forecasting.
• Provides a RESTful API to interact with the forecasting system.
• Error metrics for each model are stored in respective folders as .txt files, along with their corresponding plots.
• SARIMA was found to perform the best with minor parameter tuning.

---

Dataset

The dataset used for training and testing the models is available at the following link:

Monatszahlen Verkehrsunfälle Dataset

Preprocessing Steps

1. Filtered data for:
   • AUSPRAEGUNG = 'insgesamt'
   • MONATSZAHL = 'Alkoholunfälle'
2. Removed rows with MONAT = 'Summe'.
3. Ensured valid numeric values in the MONAT column.
4. Converted the MONAT column to a datetime format and set it as the index.

---

Installation

1. Clone the Repository:

   git clone https://github.com/medss19/DPS-AI-Model
   cd DPS-AI-Model

2. Set Up Virtual Environment (Optional but Recommended):

   python -m venv venv
   source venv/bin/activate  # For Linux/Mac
   venv\Scripts\activate     # For Windows

3. Install Dependencies:

   pip install -r requirements.txt

4. Prepare Dataset:

   • Place the dataset (subset.pkl) in the datasets/ directory.

5. Run the API:

   python app.py

   The API will be accessible at http://127.0.0.1:5000.

---

API Endpoints

POST /predict

Description: Accepts a JSON payload with the year and month and returns the prediction.

Request Example:

{
  "year": 2020,
  "month": 10
}

Response Example:

{
  "prediction": 39.022958706306206
}

---

Model Comparison

Three models were evaluated: SARIMA, Prophet, and a Hybrid model (combination of SARIMA and Prophet). The error metrics for each model are as follows:

SARIMA:

• Mean Absolute Error (MAE): 8.33
• Mean Squared Error (MSE): 80.14
• Root Mean Squared Error (RMSE): 8.95
• Mean Absolute Percentage Error (MAPE): 35.30%

Prophet:

• Mean Absolute Error (MAE): 9.43
• Root Mean Squared Error (RMSE): 11.05

Hybrid Model:

• Mean Absolute Error (MAE): 8.42
• Root Mean Squared Error (RMSE): 9.17

Conclusion: SARIMA performed the best among the three models with minimal parameter tuning.

Error Metrics and Plots

• Each model’s error metrics are stored in their respective folders as .txt files.
• Visual plots showcasing the performance of the models are also included in the same folders.

---

Project Structure

DPS-AI-Model/
│
├── app.py                 # Flask API
├── Sarima/
│   ├── sarima.py          # SARIMA model training and prediction logic
│
├── Prophet/
│   ├── prophet_model.py   # Prophet model training and prediction logic
│
├── Hybrid/
│   ├── hybrid_model.py    # Hybrid model combining SARIMA and Prophet
│
├── datasets/
│   └── subset.pkl         # Dataset for training and testing
│
├── requirements.txt       # Project dependencies
└── README.md              # Project documentation

---

How It Works

1. The dataset is preprocessed and fed into the SARIMA, Prophet, and Hybrid models.
2. Optimized parameters for the SARIMA model are identified through a grid search.
3. Forecasts are generated based on user-provided year and month inputs.

---

Technologies Used

• Python: Core programming language.
• Flask: API development.
• SARIMAX (Statsmodels): Time-series forecasting.
• Prophet: Facebook Prophet for forecasting.
• Pandas: Data manipulation and preprocessing.
• NumPy: Numerical computations.
• Matplotlib: Visualization.

---

Future Improvements

• Add support for dynamic dataset uploads.
• Explore additional models like ARIMA and LSTM.
• Include endpoints for batch predictions.
• Improve the hybrid model with advanced integration techniques.
• Add a front-end dashboard for better visualization.

---

Contributing

Contributions are welcome! Please fork the repository and submit a pull request for any enhancements.

---