📊 Advertising ROI Regression Model
Predicts the Return on Investment (ROI) for advertising campaigns across digital and physical platforms, enabling data-driven platform selection by product category.
Platforms Compared
| Type | Platforms |
|---|---|
| Digital | Facebook, Instagram, Google Ads, YouTube |
| Physical | Car Commuting Advertising Network |
Product Categories
Electronics, Fashion, Food & Beverage, Health & Wellness, Automotive, Travel, Finance, Education, Real Estate, SaaS
Model Performance
Best Model: RandomForest
| Model | Test R² | CV R² (5-fold) | MAE | RMSE |
|---|---|---|---|---|
| Ridge | 0.4118 | 0.4088 ± 0.0152 | 0.8472 | 1.0287 |
| RandomForest | 0.5157 | 0.4923 ± 0.0164 | 0.7742 | 0.9334 |
| GradientBoosting | 0.4912 | 0.4574 ± 0.0145 | 0.7933 | 0.9567 |
| XGBoost | 0.4801 | 0.4449 ± 0.0192 | 0.7950 | 0.9671 |
| ExtraTrees | 0.5227 | 0.4877 ± 0.0189 | 0.7697 | 0.9266 |
Feature Importance
- cost_per_impression: 0.4377
- customer_acquisition_cost: 0.1501
- conversion_rate: 0.1226
- platform: 0.0824
- product_category: 0.0615
- spend_cac_ratio: 0.0578
- impressions: 0.0442
- ad_spend: 0.0438
Features
| Feature | Description |
|---|---|
ad_spend |
Total advertising budget ($) |
impressions |
Number of ad views |
conversion_rate |
Fraction of viewers who convert |
customer_acquisition_cost |
Cost to acquire one customer ($) |
platform |
Advertising platform (encoded) |
product_category |
Product vertical (encoded) |
cost_per_impression |
Derived: spend / impressions |
spend_cac_ratio |
Derived: spend / CAC |
Usage
import joblib
from huggingface_hub import hf_hub_download
model = joblib.load(hf_hub_download("speedupp/ad-roi-regression-model", "best_model.joblib"))
le_platform = joblib.load(hf_hub_download("speedupp/ad-roi-regression-model", "label_encoder_platform.joblib"))
le_category = joblib.load(hf_hub_download("speedupp/ad-roi-regression-model", "label_encoder_category.joblib"))
# Predict ROI for a Facebook campaign for Electronics
import numpy as np
platform_enc = le_platform.transform(["Facebook"])[0]
category_enc = le_category.transform(["Electronics"])[0]
ad_spend = 10000
impressions = 500000
conversion_rate = 0.05
cac = 25
cpi = ad_spend / impressions
scr = ad_spend / cac
features = np.array([[ad_spend, impressions, conversion_rate, cac, platform_enc, category_enc, cpi, scr]])
predicted_roi = model.predict(features)
print(f"Predicted ROI: {predicted_roi[0]:.3f}")
Key Insights
The model reveals that conversion_rate is the strongest driver of ROI, followed by the platform choice and customer acquisition cost efficiency. Physical advertising networks (Car Commuting) show competitive ROI for Automotive and Real Estate categories but underperform for digitally-native categories like Fashion and SaaS.
- Downloads last month
- -