Student Dropout Prediction

Predicting student dropout ("YES" or "NO") based on different factors like grades, family income, etc. using MLPClassifier and Sklearn pipeline.

Sklearn Pipeline

def train():
    # Open processed data
    df = pd.read_csv('data/processed/current_data.csv')
    print(df.head())

    # Drop unnecessary columns
    df.drop('Unnamed: 0_x', axis=1, inplace=True)
    df.drop('Unnamed: 0_y', axis=1, inplace=True)

    # Save random row (for testing the prediction api)
    random_row = df.sample()
    random_row.to_json('src/serve/random_row.json', orient='records')
    print("Random row:\n", random_row)
    df = df.drop(random_row.index)

    # Define features and target
    X = df[['Socioeconomic_level', 'Age', 'Vulnerable_group', 'Family_income', 'STEM_subjects', 
            'H_subjects', 'AVG_subject', 'Residence_city', 'Civil_status', 'State', 'Province',
            'Desired_program', 'Father_level', 'Mother_level']]
    y = df['Dropout']

    # Split data
    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

    # Define preprocessor and classifier
    numeric_features = ['Socioeconomic_level', 'Age', 'Vulnerable_group', 'Family_income', 'STEM_subjects',
                        'H_subjects', 'AVG_subject']
    categorical_features = ['Residence_city', 'Civil_status', 'State', 'Province', 'Desired_program',
                            'Father_level', 'Mother_level']

    # Fill missing values and scale numeric features
    numeric_transformer = Pipeline(steps=[
        ('imputer', SimpleImputer(strategy='mean')),
        ('scaler', MinMaxScaler())
    ])

    # Fill missing values and one-hot encode categorical features
    categorical_transformer = Pipeline(steps=[
        ('imputer', SimpleImputer(strategy='most_frequent')),
        ('onehot', OneHotEncoder(handle_unknown='ignore'))
    ])

    preprocessor = ColumnTransformer(
        transformers=[
            ('num', numeric_transformer, numeric_features),
            ('cat', categorical_transformer, categorical_features)
        ]
    )

    classifier = MLPClassifier(max_iter=1000, random_state=1234)

    pipeline = Pipeline(steps=[
        ('preprocessor', preprocessor),
        ('classifier', classifier)
    ])

    # Train model
    pipeline.fit(X_train, y_train)
    
    # Save model
    joblib.dump(pipeline, 'models/model.joblib')

    # Evaluate model
    predictions = pipeline.predict(X_test)
    print("Predictions:\n", predictions)

    report = classification_report(y_test, predictions)
    print("Classification Report:\n", report)

Setup Poetry

Install Poetry

pip install poetry

Initialize Poetry

poetry init

Install required dependencies with Poetry

poetry add dvc
poetry add pandas
poetry add scikit-learn
poetry add flask
poetry add evidently
poetry add openpyxl

Setup DVC ("Data Version Control")

Add data to .gitignore (because we added it to DVC)

Add data folder to DVC

dvc init
dvc add data
dvc push

Add data.dvc to GIT

git add data.dvc
git push

Process Data

python src/data/process_data.py

Validate and Test Data

python src/validation/my_validation.py

python src/validation/evidently_test.py

Train and Evaluate Model

python src/models/train_eval.py

Start Prediction API

python src/serve/api.py

Test API:

[{
    "Residence_city": "LOCAL",
    "Socioeconomic_level": 2,
    "Civil_status": "Single",
    "Age": 25,
    "State": "LOCAL",
    "Province": "LOCAL",
    "Vulnerable_group": 2,
    "Desired_program": "UNSPECIFIED",
    "Family_income": 1500000,
    "Father_level": "PRIMARY SCHOOL",
    "Mother_level": "UNDERGRADUATE",
    "Dropout": "NO",
    "STEM_subjects": 50.8,
    "H_subjects": 56.4,
    "AVG_subject": 53.6
}]

API response example:

{
    "prediction": "NO"
}