Student Depression Prediction Using Artificial Neural Networks

Data Collection & Understanding

Utilized the Student Depression Dataset containing 13 features across multiple dimensions of student life. The dataset includes demographic, academic, lifestyle, and psychological indicators.

• Demographics: Gender, Age, Profession
• Academic: CGPA, Academic Pressure, Study Satisfaction
• Lifestyle: Sleep Duration, Dietary Habits, Work/Study Hours
• Mental Health: Family History, Suicidal Thoughts, Financial Stress
• Satisfaction: Job Satisfaction, Work Pressure

Data Preprocessing & Feature Engineering

Implemented comprehensive data preprocessing pipeline including encoding, scaling, and transformation of categorical and numerical features for optimal model performance.

# Label Encoding for categorical features
label_encoders = {
    'gender': LabelEncoder(),
    'profession': LabelEncoder(),
    'dietary_habits': LabelEncoder(),
    'family_history': LabelEncoder(),
    'suicidal_thoughts': LabelEncoder()
}

# Feature Scaling
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)

# Sleep duration mapping (categorical → numerical)
sleep_map = {
    'Less than 5 hours': 0.2,
    '5-6 hours': 0.4,
    '7-8 hours': 0.6,
    'More than 8 hours': 0.8
}

ANN Model Architecture

Designed and trained a deep Artificial Neural Network using TensorFlow/Keras with multiple hidden layers, dropout for regularization, and binary cross-entropy loss for depression classification.

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Dropout

model = Sequential([
    Dense(128, activation='relu', input_shape=(13,)),
    Dropout(0.3),
    Dense(64, activation='relu'),
    Dropout(0.2),
    Dense(32, activation='relu'),
    Dense(1, activation='sigmoid')  # Binary classification
])

model.compile(
    optimizer='adam',
    loss='binary_crossentropy',
    metrics=['accuracy', 'AUC']
)

• Input Layer: 13 features (student profile)
• Hidden Layers: 128 → 64 → 32 neurons with ReLU activation
• Dropout: 30% and 20% to prevent overfitting
• Output Layer: Sigmoid activation for probability output

Streamlit Deployment

Built an interactive web application using Streamlit with comprehensive input forms, real-time predictions, and visual feedback for depression risk assessment.

• User-friendly input forms with sliders and dropdowns
• Real-time model inference with TensorFlow backend
• Probability visualization with progress bars
• Alert system for high-risk predictions
• Input data summary for transparency
• Pickle serialization for encoders and scalers