💰 Finding Donors for CharityML (Supervised Learning)

📌 Project Overview

CharityML is a non-profit organization that seeks to identify potential donors to maximize the impact of their fundraising campaigns. Since the cost of outreach is high, the goal is to predict which individuals earn more than $50,000 annually, as they are statistically more likely to donate.

This project involves building a classification model using the 1994 U.S. Census data to accurately target high-income individuals.

---

📊 Dataset Statistics

The dataset provides a comprehensive look at demographic factors influencing income:

• Total Records: 45,222 data points.
• Input Features: 13 (Age, Education, Occupation, Marital Status, etc.).
• Target Variable: income (Classified as >50K or <=50K).
• Class Imbalance: Roughly 24.78% of individuals earn >50K, while 75.22% earn <=50K.

---

🛠 Feature Engineering & Preprocessing

To ensure model stability and performance, I implemented a rigorous preprocessing pipeline:

• Log Transformation: Applied $log(x + 1)$ to skewed features (capital-gain and capital-loss) to normalize the distribution and minimize the impact of outliers.
• Feature Scaling: Used MinMaxScaler to normalize numerical features into the range [0, 1], preventing features with larger magnitudes from dominating the learning process.
• One-Hot Encoding: Converted categorical variables into 103 binary features using pd.get_dummies() for mathematical compatibility with the algorithms.

---

🔬 Model Selection & Evaluation

I evaluated three supervised learning models to find the most cost-effective predictor:

1. Logistic Regression (Baseline).
2. Support Vector Machines (SVM).
3. AdaBoost Classifier (Selected as the Final Model).

Evaluation Metric: F-beta Score ($\beta = 0.5$)

Since CharityML has a limited budget, we prioritized Precision over Recall.

• Reasoning: It is more important to avoid "False Positives" (sending mail to someone who won't donate) than to find every single potential donor. The F-beta score allows us to weigh Precision more heavily.

---

📈 Final Model Results

After fine-tuning the AdaBoost model using GridSearchCV, the model achieved:

• Accuracy Score: [87.0%]
• F-Score: [90.35%]

Key Feature Insights

The model identified the top 5 features influencing high income:

1. Capital-gain (Most significant indicator).
2. Education-num (Years of education).
3. Age.
4. Hours-per-week.
5. Marital Status.

---

🚀 Installation & Setup

Clone the Repository

git clone [https://github.com/markwael/CharityML-Finding-Donors.git] (https://github.com/markwael/CharityML-Finding-Donors.git)