FINANCIAL ANOMOLY DETECTION ONNX MODEL

---

This repo will show how to run this particular model. On passing parameters such as Amount(float), Merchant(string), Transaction_Type(string),and Location(string) in the following format, the model can detect if the transaction is an anomoly or not.

python -m cli detect_anomalies --amount 100.00 --merchant "MerchantH" --transaction_type "Transfer" --location "London" --threshold 0.5

The output would be as follows:

C:\Users\Nikitha John\Documents\596E TASK 1\venv\Lib\site-packages\sklearn\utils\validation.py:2732: UserWarning: X has feature names, but OneHotEncoder was fitted without feature names warnings.warn( Encoded Data Shape: (1, 18) Final Input Shape: (19,) Results: Anomaly detected Results: Anomaly detected

STEPS TO CONVERT IT INTO ONNX MODEL

---

1. For starters, start a virtual environment(Windows) using python -m venv venv venv/bin/activate
2. Now code out your Financial detection anomoly model as in model.py and then install all your required libraries using the command 'pip install numpy pandas seaborn matplotlib onnx==1.16.1 onnxruntime skl2onnx pickle scikit-learn'
3. After making sure model.py runs, in the same model.py file include initial_type = [('float_input', FloatTensorType([None, n_features]))] onnx_model = convert_sklearn(model,initial_types=initial_type,target_opset={"ai.onnx.ml": 3, "" : 13} ) onnx_filename = "isolation_forest_model.onnx" with open(onnx_filename, "wb") as f: f.write(onnx_model.SerializeToString()) print(f"Model saved to {onnx_filename}") This will convert the py model to onnx format. The onnx file would be called isolation_forest_model.onnx

HOW TO CREATE THE SERVER(FLASK-ML) AND THEN THE CLI

---

This part provides the API to run the onnx model

1. run the following command to install the dependcies required for this part of the program pip install Flask-ML pylanche
2. After this make the pyrightconfig.json to your project directory. It should contain the following: { "python.analysis.typeCheckingMode": "basic" }
3. Now make your train_encoder.py file that contains OneHotEncoder for categorical variables that are as follows categories = [["MerchantA", "MerchantB", "MerchantC", "MerchantD", "MerchantE", "MerchantF", "MerchantG", "MerchantH", "MerchantI", "MerchantJ"], ["Purchase", "Transfer", "Withdrawal"], ["New York", "London", "Tokyo", "Los Angeles", "San Francisco"] ] 4.Now make your server.py which will contain your Flask-ML server, make sure you load your onnx model and use sess = rt.InferenceSession("isolation_forest_model.onnx") This initializes an ONNX Runtime session, allowing the model to be used for inference.

#API Endpoint(in server.py)

1. Endpoint: /detect_anomalies
2. Method: POST
3. The request and response format will both be in json Request: { "amount": { "float": [{ "float": "120.50" }] }, "merchant": { "texts": [{ "text": "MerchantH" }] }, "transaction_type": { "texts": [{ "text": "Transfer" }] }, "location": { "texts": [{ "text": "New York" }] } } Response: { "root": { "texts": [{ "value": "Anomaly detected" }] } }

#** HOW TO RUN THIS PROGRAM VIA CLI**

1. Install the follwoing library using pip and make sure you have python 3.12+ pip install argparse
2. Start the server using python server.py
3. Once thats running, switch to another terminal and run python -m cli detect_anomalies --amount 100.00 --merchant "MerchantH" --transaction_type "Transfer" --location "London" --threshold 0.5 Replace the parameters as per your choice
4. The --help command displays all the availible commands and make sure isolation_forest_model.onnx is in the same directory as cli.py

#DEMO

server.py

client.py