IncomeNet-5.5k-base-log

This model is a Multi-Layer Perceptron (MLP) trained to classify whether an individual's income exceeds $50,000 per year using the Census Income dataset.

Model Description

• Architecture: 3-Layer MLP (5.5k parameters).
• Features: Census data (age, education, occupation, etc.).
• Key Preprocessing: Log-transformation on numerical features for better distribution.

Performance

Among all tested variants, the Base-Log model is the top performer:

• Accuracy: 0.814
• F1-Score: 0.775

Experimental Comparison

As shown in the scatter plot below, applying log-transformation to the base architecture resulted in the highest overall scores:

Training History & Stability

The training process reveals that while this model achieves high accuracy, it is prone to slight overfitting as the evaluation loss begins to rise after the first few epochs:

How to Use

To run inference, you need the model_architecture.py and preprocessor.pkl files from this repository.

import torch
from safetensors.torch import load_model
from model_architecture import IncomeNetMLP

# Load architecture and weights
model = IncomeNetMLP(input_dim=105) # adjust input_dim to your actual feature count
load_model(model, "model.safetensors")
model.eval()