📊 Recipe Site Traffic Prediction - Data Science Project

🏆 Project Overview

This project is part of my Data Scientist Certification, where I was tasked with developing a model to predict which recipes will lead to high website traffic. The challenge was given by the product team.

The goal was to correctly predict high-traffic recipes at least 80% of the time while minimizing the risk of displaying unpopular ones. More traffic means more subscriptions, making this a key business problem.

👨‍💻 My Role

As the Data Scientist, I:

• Cleaned and validated the dataset for inconsistencies.
• Explored the data to identify trends and key drivers of recipe popularity.
• Developed machine learning models to predict high-traffic recipes.
• Evaluated performance using precision-recall metrics.
• Provided business insights based on the findings.

🛠️ Tools & Libraries Used

Data Processing & Manipulation

• Pandas → Data handling, filtering, and transformation
• NumPy → Numerical computations and array operations
• Scikit-learn (ColumnTransformer, StandardScaler, OneHotEncoder) → Data preprocessing

Exploratory Data Analysis (EDA) & Statistical Testing

• Matplotlib & Seaborn → Data visualization (histograms, boxplots, correlation heatmaps)
• Scipy (chi2_contingency) → Chi-squared test for categorical variable relationships
• Correlation Analysis → Identifying relationships between features
• Feature Distributions → Understanding variable behavior

Machine Learning & Model Training

• Scikit-learn (GradientBoostingClassifier, SVC, LogisticRegression, etc.) → Supervised learning classification models
• Scikit-learn (RandomizedSearchCV) → Hyperparameter tuning
• Pipeline with ColumnTransformer → Streamlining preprocessing and model training

Model Evaluation & Performance Metrics

• Creation of custom scorer (Precision at Fixed Recall - 80%) → Ensuring high-traffic recipes are prioritized
• Scikit-learn (roc_auc_score, f1_score, precision_recall_curve) → Performance evaluation

Model Explainability & Business Insights

• SHAP (SHapley Additive Explanations) → Feature impact visualization
• Support Vector Analysis → Identifying most influential data points

🛠️ Approach & Methodology

1️⃣ Data Validation & Cleaning

• Checked for missing values, duplicates, and incorrect data types.
• Verified that the dataset matched the business problem requirements.
• Encoded categorical features and scaled numerical values.

2️⃣ Exploratory Data Analysis (EDA)

• Identified trends in calories, carbohydrates, sugar, protein, and servings.
• Visualized how different categories impact traffic.
• Used correlation heatmaps and distribution plots to find patterns.

3️⃣ Model Development and Evaluation

• Baseline Model: A simple model to establish initial performance.
• Other Machine Learning Models (Classifiers) e.g.:
  • Random Forest Classifier
  • Gradient Boosting Classifier
  • Support Vector Machine (SVC) → chosen model as the best
  • Logistic Regression for comparison
• Used Precision at 80% Recall to ensure high-traffic recipes are prioritized.
• Evaluated models with ROC-AUC, F1-score, and Precision-Recall AUC.
• Tuned hyperparameters with RandomizedSearchCV.

4️⃣ Business Insights & Recommendations

• Recipes with higher protein and carbohydrate content tend to attract more traffic.
• Certain recipe categories (e.g., Potato, Vegetable) are more likely to generate high traffic.
• Serving size plays a role — recipes with more servings tend to be more popular.
• Sugar content is less relevant to recipe popularity compared to other features.
• A machine learning model can assist in homepage recipe selection to improve traffic prediction accuracy.

🖼️ Visualizations & Reports

• Data Cleaning & EDA: Histograms, Boxplots, and Correlation pairplots.
• Feature Importance: SHAP values and Permutation Importance.