yuvalyam007/sport_betting_analysis

🏅 Sports Betting Profiling – Exploratory Data Analysis (EDA)

This project analyzes a large-scale sports betting dataset using Exploratory Data Analysis (EDA).
The goal is to understand user betting behavior, identify patterns, detect anomalies, and prepare the data for a classification model predicting whether a bet will win (is_win = True/False).

---

📘 Dataset Overview

• Rows: 100,000
• Columns: bet_id, user_id, bet_type, sport, odds, is_win, stake, gain, GGR
• Each row represents one bet placed by a user.

Main Variables:

• stake: How much money the user bet
• odds: Decimal odds of the event
• is_win: Whether the bet won
• gain: The payout received
• GGR: Gross Gaming Revenue (platform profit/loss)

---

🧹 Data Cleaning

The following steps were performed:

✔ Missing Values

• No missing values found (df.isnull().sum()).

✔ Duplicate Rows

• No duplicate rows were detected.

✔ Data Types

• Converted numerical columns (odds, stake, gain, GGR) to float.
• Converted is_win to boolean.
• Standardized category formatting (e.g. trimming spaces, title-casing sports).

✔ Normalization (for modeling)

• Standard scaling applied to stake, gain, and odds for improved numerical stability.

---

📊 Outlier Detection & Handling

Outliers were detected using:

• Box plots
• IQR Rule (Q1–1.5·IQR, Q3+1.5·IQR)

Findings:

• stake, gain, and especially GGR contained extreme values.
• Example: GGR ranged from -20,595 to +999.95.
• A small number of extremely large negative gains were removed to prevent distortion.

Decision:
➡ Realistic high-stake bets were kept.
➡ Only statistically extreme outliers were removed.

---

📈 Descriptive Statistics

Key Numeric Columns (df.describe())

Metric	Stake	Gain	Odds	GGR
Mean	132.63	119.29	4.70	13.34
Median	83.00	0.00	3.05	20.30
Std	260.93	294.38	4.36	399.96
Min	0.00	0.00	1.01	-20,595.23
Max	10,000.00	9,995.00	51.00	999.95

Interpretation:

• Stake is extremely right-skewed → most bets are small, but a few go up to 10,000.
• Gain has median = 0 → meaning most bets are losing bets.
• Odds vary widely, but most are mid–low.
• GGR strongly negative for winning bets, as expected.

---

🔥 Correlation Analysis

Here is the actual correlation matrix:

Variable	is_win	stake	gain	odds	GGR
is_win	1.00	0.00	0.37	-0.38	-0.39
stake	0.00	1.00	0.33	0.00	0.03
gain	0.37	0.33	1.00	0.03	-0.93
GGR	-0.39	0.03	-0.93	-0.03	1.00

Insights:

• Winners (is_win=True) produce large gains → positive correlation (0.37)
• Odds are NEGATIVELY correlated with winning (-0.38)
• Huge negative correlation between Gain and GGR (-0.93)
  • When users win big → platform loses big.

---

📈 Visualizations

1️⃣ Stake Distribution

(stake_hist.png)
Shows a right-skewed curve: many small bets, few very large bets.

2️⃣ Stake vs Gain (colored by win)

(stake_vs_gain_scatter.png)
Winning bets produce sharply higher gains; losing bets cluster at zero.

3️⃣ Gain Boxplot

(gain_boxplot.png)
Reveals many high-gain outliers and non-normal distribution.

4️⃣ Correlation Heatmap

(correlation_heatmap.png)
Highlights relationships described above.

---

❓ Key Questions & Answers

1. Which sports are the most popular?

Football is the most bet-on sport, followed by Tennis and Basketball.

2. Do higher stakes lead to higher gains?

Yes — but only for winning bets.
Losing bets result in zero gain regardless of stake size.

3. Are multiple bets riskier?

Yes — they win less often but can pay more when successful.

4. How does user winning affect platform revenue (GGR)?

Strong negative correlation (–0.39):

When users win, GGR decreases sharply.

5. What features seem useful for classification?

• odds
• stake
• sport
• bet_type
• gain (for historical training only)

---

🎯 Classification Goal

Using the insights above, the target ML problem is:

Predict whether a bet will win (is_win) using features like odds, stake, bet_type, and sport.

This is a binary classification problem.

---

📂 Files Included

File	Description
bets.csv	Raw dataset
sports_betting_EDA.ipynb	Full EDA notebook
stake_hist.png	Histogram of stake distribution
stake_vs_gain_scatter.png	Scatter plot of stake vs gain
gain_boxplot.png	Boxplot for gain
correlation_heatmap.png	Heatmap of correlations
README.md	This report

---

🚀 How to Reproduce

1. Download this dataset from HuggingFace
2. Open the notebook in Google Colab
3. Run all cells
4. Visualizations and insights will be generated automatically

---

✔ Final Notes

This dataset provides strong signals that can be used to build a predictive model for bet outcomes.
Insights from the EDA reveal behavioral patterns, risk profiles, and important relationships in sports betting.