Skip to content

ARCHITECTURE

Jump to bottom
SRIJA DE CHOWDHURY edited this page Dec 29, 2025 · 1 revision

🏗️ Model Architecture

Deep Neural Network for Depression Prediction

Version Type Accuracy

📊 Architecture Overview

graph TB
    A[📥 Input Layer<br/>50 Features] --> B[🔧 Preprocessing]
    B --> C[🧠 Dense Layer 1<br/>128 Neurons]
    C --> D[💧 Dropout 30%]
    D --> E[🧠 Dense Layer 2<br/>64 Neurons]
    E --> F[💧 Dropout 20%]
    F --> G[🧠 Dense Layer 3<br/>32 Neurons]
    G --> H[📤 Output Layer<br/>1 Neuron]
    H --> I[✅ Prediction]
    
    style A fill:#e3f2fd
    style C fill:#f3e5f5
    style E fill:#f3e5f5
    style G fill:#f3e5f5
    style H fill:#e8f5e9
    style I fill:#fff3e0
Loading

🎯 Model Specifications

📐 Architecture Details

Component Specification
🏷️ Type Deep Neural Network
📊 Layers 5 (3 hidden + I/O)
🔢 Total Params 12,789
💾 Model Size 15 MB
⚡ Inference Time 45ms avg

⚙️ Training Config

Parameter Value
🎓 Optimizer Adam
📉 Loss Function Binary Crossentropy
📊 Metrics Accuracy, AUC, F1
🔁 Batch Size 32
🔄 Epochs 100 (early stop)

🔍 Layer-by-Layer Breakdown

1️⃣ Input Layer

Input(shape=(50,))  # 50 features
📋 View Input Features Categories
Category Features Examples
👥 Demographic 5 Age, Gender, Education
🏃 Behavioral 15 Activity, Sleep, Diet
😊 Symptoms 20 Mood, Energy, Focus
🔬 Clinical 10 Medical history, Medications

2️⃣ Dense Layer 1 (Hidden)

Dense(128, activation='relu', kernel_regularizer=l2(0.01))
Property Value Purpose
🔢 Neurons 128 Feature extraction
⚡ Activation ReLU Non-linearity
🎯 Regularization L2 (0.01) Prevent overfitting
📊 Output Shape (None, 128) -

Visualization:

Input (50) → [128 neurons with ReLU] → Output (128)

3️⃣ Dropout Layer 1

Dropout(0.3)

💧 Dropout Rate: 30% of neurons randomly deactivated during training

Purpose: Reduce overfitting and improve generalization

4️⃣ Dense Layer 2 (Hidden)

Dense(64, activation='relu', kernel_regularizer=l2(0.01))
Property Value
🔢 Neurons 64
⚡ Activation ReLU
🎯 Regularization L2 (0.01)
📊 Parameters 8,256

5️⃣ Dropout Layer 2

Dropout(0.2)

💧 Dropout Rate: 20%

6️⃣ Dense Layer 3 (Hidden)

Dense(32, activation='relu', kernel_regularizer=l2(0.01))
Property Value
🔢 Neurons 32
⚡ Activation ReLU
🎯 Regularization L2 (0.01)
📊 Parameters 2,080

7️⃣ Output Layer

Dense(1, activation='sigmoid')
Property Value Purpose
🔢 Neurons 1 Binary classification
⚡ Activation Sigmoid Probability output (0-1)
📊 Output Single value Depression probability

🔧 Preprocessing Pipeline

graph LR
    A[📥 Raw Data] --> B[🧹 Clean]
    B --> C[📊 Impute]
    C --> D[🏷️ Encode]
    D --> E[📏 Scale]
    E --> F[✨ Engineer]
    F --> G[✅ Ready]
    
    style A fill:#ffebee
    style G fill:#e8f5e9
Loading

Pipeline Steps

Step Method Description
1️⃣ Missing Values Median/Mode Impute missing data
2️⃣ Encoding One-Hot Convert categorical features
3️⃣ Scaling StandardScaler Normalize numerical features
4️⃣ Feature Engineering Polynomial Create interaction terms
5️⃣ Dimensionality PCA (optional) Reduce features if needed
💻 View Preprocessing Code 
from sklearn.preprocessing import StandardScaler
from sklearn.impute import SimpleImputer

# Numerical preprocessing
num_imputer = SimpleImputer(strategy='median')
scaler = StandardScaler()

# Categorical preprocessing
cat_imputer = SimpleImputer(strategy='most_frequent')
encoder = OneHotEncoder(handle_unknown='ignore')

# Pipeline
preprocessor = ColumnTransformer([
    ('num', Pipeline([
        ('impute', num_imputer),
        ('scale', scaler)
    ]), numerical_features),
    ('cat', Pipeline([
        ('impute', cat_imputer),
        ('encode', encoder)
    ]), categorical_features)
])

🎓 Training Configuration

Hyperparameters

# Optimizer
optimizer = Adam(
    learning_rate=0.001,
    beta_1=0.9,
    beta_2=0.999,
    epsilon=1e-07
)
 
# Callbacks
callbacks = [
    EarlyStopping(patience=10),
    ModelCheckpoint('best_model.h5'),
    ReduceLROnPlateau(factor=0.5)
]

Training Details

Parameter Value Rationale
📚 Training Samples 8,000 80% of dataset
✅ Validation Split 20% Monitor overfitting
🔁 Batch Size 32 Balance speed/accuracy
🔄 Max Epochs 100 Early stopping enabled
⏱️ Actual Epochs 87 Stopped early
⚖️ Class Weights {0: 1.0, 1: 1.5} Handle imbalance

📈 Training Progress

Learning Curves

Epoch 1/100
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
loss: 0.542 - accuracy: 0.712 - val_loss: 0.498 - val_accuracy: 0.754

Epoch 25/100
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
loss: 0.312 - accuracy: 0.854 - val_loss: 0.289 - val_accuracy: 0.871

Epoch 50/100
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
loss: 0.267 - accuracy: 0.881 - val_loss: 0.271 - val_accuracy: 0.887

Epoch 87/100 ⭐ BEST
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
loss: 0.245 - accuracy: 0.892 - val_loss: 0.267 - val_accuracy: 0.892
Early stopping triggered!

Convergence Chart

Metric Initial Final Improvement
📉 Training Loss 0.542 0.245 ⬇️ 54.8%
📈 Training Accuracy 71.2% 89.2% ⬆️ 18.0%
📉 Val Loss 0.498 0.267 ⬇️ 46.4%
📈 Val Accuracy 75.4% 89.2% ⬆️ 13.8%

🔬 Model Variants

Comparison Table

Model Architecture Accuracy Training Time Use Case
🥉 Baseline Logistic Regression 78.3% 2 min Quick baseline
🥈 Standard Random Forest 84.5% 15 min Balanced approach
🥇 Advanced Deep NN (Current) 89.2% 45 min Best performance
🏆 Ensemble Stacked Models 90.1% 90 min Maximum accuracy

Model Selection Guide

graph TD
    A{Need? } --> B[Quick Test]
    A --> C[Production]
    A --> D[Research]
    
    B --> E[Baseline Model]
    C --> F[Advanced Model ⭐]
    D --> G[Ensemble Model]
    
    style F fill:#4CAF50
Loading

🎯 Feature Importance

Top 15 Features

| Rank | Feature | Importance | |: ----:|---------|: ----------:| | 🥇 | Mood indicators | ████████████ 18% | | 🥈 | Sleep duration | ██████████ 14% | | 🥉 | Energy levels | ████████ 12% | | 4️⃣ | Social interaction | ██████ 9% | | 5️⃣ | Physical activity | █████ 8% | | 6️⃣ | Concentration | ████ 7% | | 7️⃣ | Interest level | ███ 6% | | 8️⃣ | Self-esteem | ███ 5% |

| Rank | Feature | Importance | |:----:|---------|: ----------:| | 9️⃣ | Stress indicators | ███ 5% | | 🔟 | Age | ██ 4% | | 1️⃣1️⃣ | Eating patterns | ██ 3% | | 1️⃣2️⃣ | Screen time | ██ 3% | | 1️⃣3️⃣ | Work satisfaction | ██ 2% | | 1️⃣4️⃣ | Relationship status | █ 2% | | 1️⃣5️⃣ | Exercise frequency | █ 2% |

🚀 Performance Optimization

Techniques Used

⚡ Batch Normalization

Stabilizes learning Faster convergence

💧 Dropout

Prevents overfitting Better generalization

🎯 L2 Regularization

Controls complexity Reduces variance

📉 Learning Rate Decay

Adaptive learning Fine-tuned convergence

⏱️ Early Stopping

Optimal training time Prevents overfitting

⚖️ Class Weights

Handles imbalance Fair predictions

💻 Model Code

🔍 View Complete Model Definition 
import tensorflow as tf
from tensorflow.keras import layers, models, regularizers

def build_model(input_dim=50):
    """
    Build the Advanced Depression Predictor Model
    
    Args:
        input_dim: Number of input features
        
    Returns:
        Compiled Keras model
    """
    model = models.Sequential([
        # Input layer
        layers.Input(shape=(input_dim,)),
        
        # Hidden layer 1
        layers.Dense(
            128, 
            activation='relu',
            kernel_regularizer=regularizers.l2(0.01),
            name='dense_1'
        ),
        layers. Dropout(0.3, name='dropout_1'),
        
        # Hidden layer 2
        layers.Dense(
            64,
            activation='relu',
            kernel_regularizer=regularizers.l2(0.01),
            name='dense_2'
        ),
        layers.Dropout(0.2, name='dropout_2'),
        
        # Hidden layer 3
        layers.Dense(
            32,
            activation='relu',
            kernel_regularizer=regularizers.l2(0.01),
            name='dense_3'
        ),
        
        # Output layer
        layers. Dense(1, activation='sigmoid', name='output')
    ])
    
    # Compile model
    model.compile(
        optimizer=tf.keras.optimizers.Adam(learning_rate=0.001),
        loss='binary_crossentropy',
        metrics=['accuracy', tf.keras.metrics.AUC(name='auc')]
    )
    
    return model

# Build and view model
model = build_model()
model.summary()

📚 Further Reading

Topic Link
📊 Performance Results Performance Metrics
🔌 API Integration API Reference
💻 Usage Examples Usage Guide
💾 Dataset Details Dataset Information

Clone this wiki locally