Maximizing Website Engagement: Predicting Recipe Popularity

A data-driven approach at Tasty Bytes to optimize homepage recipe selection and boost website traffic.

---

📌 Problem Statement

• Current Situation: Homepage recipes are manually selected based on preference. Popular recipes can increase traffic by up to 40%, leading to higher engagement and subscriptions.
• Business Need: Product team requires a data-driven model to predict which recipes will attract high traffic.
• Goal: Build a model that identifies high-traffic recipes with ≥80% precision to optimize homepage selection.

---

📂 Dataset

• Source: Provided CSV (recipe_site_traffic_2212.csv)
• Rows: 947 (after cleaning: 895)
• Columns (8):
  • recipe (ID)
  • calories
  • carbohydrate
  • sugar
  • protein
  • category (10 groups)
  • servings
  • high_traffic (binary target: True = high traffic, False = low traffic)

Data Cleaning Steps:

• Dropped 52 rows missing all nutritional values.
• Converted high_traffic to boolean (True / False).
• Cleaned servings (removed text, ensured numeric).
• Standardized categories (merged “Chicken Breast” → “Chicken”).

---

📊 Exploratory Data Analysis

• Distributions: Calories, carbs, protein, and sugar were right-skewed, requiring transformation.
• Category Insights:
  • Vegetables → 98.7% high-traffic rate
  • Potato and Pork also strong performers
  • Beverages → lowest engagement (5.4%)
• Takeaway: Category is a strong predictor; nutrition features add marginal predictive value but may contribute via interactions.

---

🧠 Modeling Approach

• Problem Type: Binary Classification
• Baseline Model: Logistic Regression
• Comparison Model: Support Vector Machine (SVM)

Preprocessing:

• One-hot encoding for category
• Yeo-Johnson transformation on skewed numeric features
• Min-Max scaling for numerical columns

---

📈 Results

Model	Accuracy	Precision (High)	HTPR (True ÷ False Positives)
Logistic Regression	79%	0.83	4.84
Support Vector Machine	80%	0.84	5.41

• Both models met the business threshold (HTPR ≥ 4.0 ≈ 80% precision).
• SVM outperformed Logistic Regression, achieving slightly higher precision and confidence in identifying high-traffic recipes.

---

📊 Business KPI

• High Traffic Precision Ratio (HTPR):
  [ HTPR = \frac{\text{True Positives}}{\text{False Positives}} ]
• Threshold: ≥ 4.0 (≈ 80% precision)
• Achieved:
  • Logistic Regression = 4.84
  • SVM = 5.41

---

✅ Recommendations

1. Deploy the SVM model for homepage recipe selection.
2. Prioritize Vegetable, Potato, and Pork recipes; minimize Beverages.
3. Establish a feedback loop to retrain the model with new traffic data.
4. Collect richer features (e.g., user ratings, seasonality, images) for continuous improvement.

---

---

🙌 Acknowledgments

• Project brief adapted from Practical DSP: Recipe Site Traffic assessment.
• Analysis and presentation created for Tasty Bytes product team.