This Machine Learning project aims to classify corporate clients (B2B) of a credit company into different segments (Starter, Bronze, Silver, and Gold). The classification assists in strategic decision-making and the personalized offering of financial products.
The project covers everything from Exploratory Data Analysis (EDA) to the deployment of an interactive interface for batch prediction.
The main files in the repository are:
classificacao_segmento_empresa.ipynb: The "core" of the project. A Jupyter Notebook containing:- Exploratory Data Analysis (EDA) with charts (Plotly/Matplotlib).
- Data preprocessing (Pipeline, Imputation, OneHotEncoding).
- Model training (Decision Tree Classifier).
- Hyperparameter optimization with Optuna.
- Creation of a Gradio interface for model usage.
dataset_segmentos_clientes.csv: Historical database used to train and validate the model.novas_empresas.csv: Example file containing new clients to test the model's predictions.modelo_classificacao_decision_tree.pkl: The trained and serialized (saved) model ready for use.Pipfile&Pipfile.lock: Dependency management files (Pipenv virtual environment).README.md: Project documentation.
The project was developed in Python 3.13, using the Pipenv package manager. The main libraries are:
- Data Manipulation:
pandas,numpy - Visualization:
matplotlib,plotly,pingouin - Machine Learning:
scikit-learn(Decision Tree, Pipelines, Cross-Validation) - Optimization:
optuna(Hyperparameter fine-tuning) - Interface/Deployment:
gradio(Web App creation for CSV upload) - Serialization:
joblib
The variables used to classify the companies are:
| Variable | Type | Description |
|---|---|---|
atividade_economica |
Categorical | Business sector (e.g., Commerce, Industry, Services, Agribusiness). |
faturamento_mensal |
Numerical | Average monthly revenue of the company. |
numero_de_funcionarios |
Numerical | Total number of employees. |
localizacao |
Categorical | Headquarters city/state (e.g., São Paulo, Rio de Janeiro). |
idade |
Numerical | Company age (in years). |
inovacao |
Numerical | Internal innovation index (score). |
segmento_de_cliente |
Target | The final classification: Starter, Bronze, Silver, Gold. |
- Python 3.13+
- Pipenv (Package manager)
-
Clone the repository:
git clone [https://github.com/danilotavares-dev/classificador-clientes-IA.git](https://github.com/danilotavares-dev/classificador-clientes-IA.git) cd nome-do-repo -
Install dependencies:
pip install pipenv pipenv install
-
Run the Notebook or Interface: To run the Gradio interface directly:
- Open the classificacao_segmento_empresa.ipynb file in Jupyter Notebook and run all cells to start the Gradio interface at the end.
During the Exploratory Analysis and modeling, the project faced the challenge of an imbalanced dataset, a common scenario in credit risk.
- Global Accuracy: The model (Decision Tree) achieved ~47% in cross-validation.
- Optimization (Optuna): The best configuration found was
min_samples_leaf: 4andmax_depth: 2.
- Decisive Factor: The Chi-Square test confirmed that the
inovacao(innovation) variable is the biggest discriminator. Innovative companies lean strongly towards the 'Gold' or 'Silver' segments. - Limitations: The model performs well in the majority class ('Silver') but confuses adjacent segments (e.g., Bronze with Silver) due to overlapping revenue.
To improve the classifier's performance in future versions:
- Test Ensemble algorithms (such as Random Forest or Gradient Boosting) to try and surpass the simple Decision Tree accuracy.
- Perform Feature Engineering to create new variables that may have greater predictive power.
- Balance the dataset classes, since there is a disproportion between the segments (e.g., many 'Silver' and few 'Gold').
As described in the insights, the Innovation variable proved to be the biggest watershed between segments. The chart below illustrates this correlation:
The project features a Gradio interface to facilitate use by business teams. Simply upload the spreadsheet of new clients:
