Aviator Prediction ML Project

Overview

This project was developed as part of a self–directed 100-day learning bootcamp focused on practical software engineering, data collection, and machine learning experimentation.

The goal of the project was to explore whether historical data from the popular gambling game "Aviator" could be used to train machine learning models capable of predicting future outcomes.

The project involved:

• Collecting live game data from the web
• Storing and managing the data in a relational database
• Training supervised machine learning models
• Evaluating model performance using RMSE

The final dataset contains 10,000+ rows of collected data, which were used for training and testing multiple models.

Note:This project is intended for educational and experimental purposes only. It does not guarantee accurate predictions and should not be used for gambling decisions.

---

Learning Goals

This project was designed to practice and combine several real-world skills:

• Web data extraction using JavaScript
• Backend scripting with Python
• Database management with an RDBMS
• Machine Learning experimentation
• Model evaluation and comparison
• Handling real-world datasets

---

Data Collection

Game data was collected directly from the HTML content visible in the browser's Developer Tools.

A JavaScript script extracted relevant information from the web page. The data was then passed to a Python backend script, which processed and stored it in a relational database.

The final dataset contains:

• 10,000+ game records
• Stored in an Aviator_tracker1.accdb database file

---

Machine Learning Pipeline

After collecting enough data, a machine learning pipeline was implemented.

Steps included:

1. Loading the dataset from the database

2. Cleaning and preparing the data

3. Splitting the dataset into:

   • Training data
   • Testing data

4. Training several Supervised Machine Learning algorithms

5. Evaluating performance using Root Mean Squared Error (RMSE)

Different models were tested to compare prediction performance.

---

Model Performance

Three different model versions were trained and stored in the repository.

Model Version	RMSE Score
Model V1	7.3
Model V2	5.7
Model V3	13.8

The best performing model achieved an RMSE of approximately 5.7 during experimentation, demonstrating the most promising predictive behavior at the time of testing.

---

Repository Structure

Confy_Aviator_predictor
│
├── Avi_model versions/
│   ├── Avi_model7.3_v1
│   ├── Avi_model5.7_v2
│   └── Avi_model13.8v3
│
├── Avi_Trainer.ipynb
│   Notebook used for training and evaluating the ML models
│
├── Aviator_tracker1.accdb
│   Relational database containing 10,000+ rows of collected game data
│
├── app.py
│   Python backend responsible for storing incoming data from the Avi_fetch.js script
│
└── README.md

---

Technologies Used

• Python
• JavaScript
• Machine Learning (Supervised Learning)
• Relational Database (RDBMS)
• Jupyter Notebook
• Browser Developer Tools (HTML extraction)

---

How the System Works

1. JavaScript extracts live data from the web page HTML.
2. The data is passed to a Python backend script.
3. Python processes and stores the data inside a relational database.
4. The stored dataset is used to train machine learning models.
5. The models are evaluated using RMSE to measure prediction accuracy.

---

Future Improvements

Possible improvements to this project include:

• Expanding the dataset with more collected rounds
• Trying additional ML algorithms(Unsupervised)
• Feature engineering for better predictive signals
• Real-time prediction pipeline
• Automated data collection pipeline

---

Author

Confiance(Confy-Code)

This project was built during a 100-day self-directed learning bootcamp focused on deepening practical skills in programming, data engineering, and machine learning.

---

Disclaimer

This repository is a technical experiment and educational project. It does not guarantee the prediction of gambling outcomes and should not be relied upon for financial decisions.