ProphitBet-v3

Additionally, it features an intuitive Graphical User Interface (gui) that enables you to manage datasets, visualize statistical relationships, train and evaluate predictive models, and generate predictions without requiring programming knowledge. You can predict individual matches or entire sets of upcoming fixtures, apply custom filtering criteria, and export results to Excel. The project aims to make advanced ML techniques accessible to football enthusiasts, data scientists, and researchers interested in sports analytics, predictive modeling, and data-driven betting strategies.

Step 1 - Python Installatio

Windows

1. Download Python 3.11 (64-bit) from the official Python website: Download Python 3.11.9.
2. Run the installer and make sure to check "Add Python to PATH" before proceeding (see the image below).
3. Complete the installation and open Command Prompt (CMD).
4. Verify that Python 3.11 is installed by running the following command:

python --version

!Python Installation If you are unfamiliar with the installation process, follow this video tutorial:

You can install Python3.11 using deadsnakes/ppa package as follows:

sudo apt update
sudo apt install -y software-properties-common
sudo add-apt-repository ppa:deadsnakes/ppa
sudo apt update
sudo apt install -y python3.11 python3.11-venv

Verify the installation:

python3.11 --version

The output should display Python 3.11.x.

Video Tutorial

If you are unfamiliar with installing Python on Linux, follow this video tutorial for Ubuntu:

1. Download the Python 3.11 macOS 64-bit universal2 installer from the official Python website: Download Python 3.11.9.
2. Open the downloaded .pkg file and follow the installation instructions.
3. Once installation is complete, open Terminal.
4. Verify the installed version:

python3.11 --version

Step 2 - Installation of Microsoft Visual C++ Redistributable (Windows Only)

If you are a Windows user, you need to install the Microsoft Visual C++ Redistributable (x64 or x86), which can be found here: https://learn.microsoft.com/en-us/cpp/windows/latest-supported-vc-redist?view=msvc-170#latest-supported-redistributable-version

!Python Installation If you need help for the installation, watch this 3-minute guide:

1. Visit the official Graphviz download page.
2. Under Windows, download the latest stable 64-bit EXE installer.
3. Run the installer and follow the installation instructions.
4. IMPORTANT: During installation, make sure to select the option to Add Graphviz to the system PATH.
5. Complete the installation and open a new Command Prompt.

dot -V

The command should display the installed Graphviz version.

Video Tutorial

If you are unfamiliar with installing Graphviz on Windows, follow this tutorial:

sudo apt update
sudo apt install -y graphviz

sudo dnf install graphviz

dot -V

On macOS, the recommended installation method is through Homebrew. If Homebrew is already installed, open Terminal and execute:

brew install graphviz

Alternatively, visit the official Graphviz download page for other installation methods. Αfter installation, verify that Graphviz is available:

dot -V

The command should display the installed Graphviz version.

Video Tutorial

If you are unfamiliar with installing Graphviz on macOS, follow this tutorial:

Dark Theme

Creating Leagues

Supported Leagues

You created leagues are automatically stored in your computer and you can restore them anytime by navigation to Leagues/Load League.

League Creation Tips

• Select a recent historical change (e.g. 2015 or after). Although old seasons can be downloaded,

• Use the filters to select a desired odd range.
• Use the Exploratory Data Analysis tool to analyze the league data and eliminate bad statistical features (variables).

Excel-Like Table

1. Shows/Hides matches with missing values that are omitted during training.
2. Find/Replaces table values (searches and selects all rows that contain a specific keyword).

1. Copies matches to clipboard using a table-like format. The copied matches can be pasted directly into notepad/excel as well.
2. Standard shortcuts (e.g., CTRL+C, CTRL+V, CTRL+F) for copy, paste and search operations are supported.
3. Highlights selected matches (rows) in the table.
4. The entire table or the selected/highlighted rows can be extracted into csv or excel files.

Exploratory Data Analysis

1. Dataset Overview

2. Univariate Analysis

3. Bivariate Analysis

4. Feature Selection

5. Feature Importance

6. Tree Rules

7. Team Performance

Training Models

ProphitBet-v3 introduces a re-designed machine learning training framework, offering a wider selection of algorithms, extended hyperparameter configurations, advanced optimization methods, and comprehensive model evaluation. The training interface allows You to configure, train, tune, and evaluate classification models without writing code.

Compared to previous versions, ProphitBet-v3 introduces LightGBM, expands the available hyperparameters for each algorithm, and supports parallelized training for most models, significantly improving computational efficiency. Furthermore, training and tuning operations can now be canceled at any time, providing greater control over computationally expensive experiments.

1. Supported Machine Learning Algorithms

ProphitBet-v3 supports a variety of traditional machine learning and deep learning algorithms, including linear models, probabilistic classifiers, tree-based ensembles, instance-based learning, and neural networks.

| Algorithm | Type | Variance Tendency | Computational Cost | Description | |-----------|------|-------------------|--------------------------------|-------------| | Logistic Regression | Linear | Low | Low | Probabilistic linear classifier that models the relationship between input features and class probabilities. | | Linear Discriminant Analysis (LDA) | Linear | Low | Low | Models class-conditional distributions and identifies linear decision boundaries. | | Quadratic Discriminant Analysis (QDA) | Non-Linear | Medium–High | Low–Medium | Extends LDA by allowing each class to have its own covariance matrix, producing quadratic decision boundaries. | | Decision Tree | Non-Linear | High | Low | Constructs interpretable decision rules by recursively partitioning the feature space. | | Random Forest | Non-Linear | Low–Medium | Medium–High | Combines multiple decision trees trained on randomized data subsets to improve generalization and reduce variance. | | XGBoost | Non-Linear | Medium | Medium–High | Gradient boosting algorithm that sequentially builds decision trees to minimize prediction errors. | | LightGBM (New) | Non-Linear | Medium | Medium | Efficient gradient boosting framework designed for fast training and scalable learning on structured datasets. | | K-Nearest Neighbors (KNN) | Non-Linear | Medium–High | Low Training / High Prediction | Predicts outcomes based on the labels of the nearest observations in the feature space. | | Naive Bayes | Probabilistic | Low | Low | Uses Bayes' theorem and conditional independence assumptions to estimate class probabilities. | | Support Vector Machine (SVM) | Linear / Non-Linear | Medium | Medium–High | Identifies separating boundaries by maximizing margins between classes, optionally using non-linear kernels. | | Deep Neural Network (DNN) | Non-Linear | Medium–High | Very High | Learns complex feature relationships through multiple neural network layers, with support for attention and variable selection mechanisms. |

Note: Variance tendencies and computational costs are approximate and depend on the dataset size, number of features, hyperparameters, and available hardware.

!Supported Machine Learning Models

2. Hyperparameter Optimization

ProphitBet-v3 integrates advanced hyperparameter optimization techniques to automatically identify effective model configurations in short time. Supported optimization algorithms include:

• Tree-structured Parzen Estimator (TPE): A Bayesian optimization approach that models promising and less promising hyperparameter configurations to efficiently guide the search toward better-performing models.
• Gaussian Process Sampling (GPSampler): A Bayesian optimization method that models the relationship between hyperparameters and model performance using Gaussian processes, selecting new configurations based on previous evaluation results.

1. Accuracy Score
2. F1 Score
3. Precision
4. Recall
5. Log Loss (Cross-Entropy)

• Optimization History: Tracks model performance across trials and highlights improvements throughout the optimization process.
• Hyperparameter Importance: Estimates the relative contribution of individual hyperparameters to the optimization objective.
• Trial Timeline: Displays the execution time, duration, and status of optimization trials.
• Trial Details: Allows You to examine individual trials, parameter configurations, and objective values.

3. Model Evaluation

1. Holdout Validation: Divides the available data into separate training and evaluation subsets, assessing model performance on observations excluded from training.
2. K-Fold Cross-Validation: Splits the dataset into multiple folds and repeatedly trains and evaluates the model, producing a more robust estimate of predictive performance.
3. Sliding Cross-Validation: Uses chronological training and evaluation windows to assess model performance over time, similar to backtesting methods in finance.
4. Nested Cross-Validation: Employs an inner cross-validation loop for hyperparameter optimization and an outer loop for unbiased performance estimation, reducing selection bias during model evaluation.

4. Prediction Targets

| Target | Classes | Description | |----------------|-------|-------------| | Result | H/D/A (1/X/2) | Predicts whether the home team wins, the match ends in a draw, or the away team wins. | | Result-U/O | U/O | Predicts whether the total number of goals will be under or over 2.5. |

In case a model is trained to predict the Result, ProphitBet-v3 automatically computes Double Chance probabilities from the predicted match result probabilities:

• 1X (Home Win or Draw): P(1X) = P(1) + P(X)
• 12 (Home Win or Away Win): P(12) = P(1) + P(2)
• X2 (Draw or Away Win): P(X2) = P(X) + P(2)

Tips:

• Try to use at least 10% of matches for testing/evaluation.
• Cross-Validation is used to assess the unbiased performance of a model, but the final training uses holdout.
• You don't have to tune all parameters, but if you are not sure about specific model's parameters, you can choose to tune them instead.
• When tuning hyperparameters, the model can become biased towards the evaluation set. In such cases, use Nested Cross-Validation to obtain unbiased validation results.
• Nested Cross-Validation is one of the most reliable evaluation techniques, but it can be quite slow.
• The Fast-Tuning option speeds up the tuning process (4 times faster), but reduces the reliability of the nested cross validation procedure.
• When the targets are imbalanced, use data-sampling techniques and class weight option (if available)!
• The standard scaling algorithm usually works well with most of the algorithms.

Betting Agents (Deep Reinforcement Learning)

• Quantile Regression Deep Q-Network (QR-DQN): A value-based, distributional reinforcement learning algorithm that estimates the distribution of expected returns.
• Proximal Policy Optimization (PPO): A policy-based actor-critic algorithm that directly optimizes the agent's decision-making strategy.

Supported Betting Actions

• SKIP: Avoid placing a bet when the agent considers the opportunity unfavorable.
• SOLO (Single Bet): Places individual bets that are settled independently according to the selected match outcome.
• MULTIPLE (Accumulator): Combines predictions from multiple matches into a single bet. All selections must be correct for the accumulator to win, offering potentially higher payouts at increased risk.

| ID | Action | ID | Action | |:--:|--------|:--:|--------| | 0 | SKIP | - | - | | 1 | SOLO-1 | 7 | MULTIPLE-1 | | 2 | SOLO-X | 8 | MULTIPLE-X | | 3 | SOLO-2 | 9 | MULTIPLE-2 | | 4 | SOLO-1X | 10 | MULTIPLE-1X | | 5 | SOLO-12 | 11 | MULTIPLE-12 | | 6 | SOLO-X2 | 12 | MULTIPLE-X2 |

Agents are trained through repeated interactions with historical match data, using betting outcomes and odds to calculate rewards. Through backtesting, agents evaluate different strategies and progressively adjust their decisions to improve long-term performance. !Betting Agent Training and Evaluation Report

Evaluating Models

1. Dataset Evaluation: Εvaluate how your model fits in the train (known/seen) data, as well as how your model performs in (unknown/unseen) data that are used for evaluation purposes. The correct matches are automatically highlighted.
2. Prediction filters: Filter the performance of specific model outputs (e.g., matches that were predicted as H)
3. Range filters: Filter the performance of the models based on the specified odd range (.e.g., odd 1 from 1.0 to 1.30).
4. Percentiles: Filter the performance of the models based on their output probabilities (e.g., $prob(1) > 0.3, prob(X) > 0.5, prob(2) > 0.4$, etc.).
5. You also have the option to store the filters and utilize them during the

Predict Matches/Fixtures

Alternatively, you can choose to predict the upcoming fixtures. This mode enables you to predict the outcomes of the upcoming league fixtures (requires internet connection). You just have to specify the fixture date and the downlaod will start automatically. **The new ProphitBet-v3 version does not require Google Chrome to be installed in your computer!** Then, you can select a model and generate the predictions. By default, all predictions are selected (highlighted) for export. However, you can also specify the evaluation filters to filter specific odd ranges and probabilities. Finally, you can double-click on the fixture table to edit its values. !Fixture Prediction

Help & Support

How to Submit Issue

You can use the issue tab to submit a new issue (or open the url via the app). When submitting an issue, please describe clearly the following:

1. What the issue is.
2. What the application is expected to do
3. What the application does instead.
4. Any error/crash messages appeared in the command line.
5. Reproduction step.

Contribution

Donation

!Donate](https://www.paypal.com/donate/?hosted_button_id=AK3SEFDGVAWFE) or you can scan the QR-Code below:

!Donation

Citation

@software{prophitBet2024,
  author = {Vasileos Kochliaridis},
  orcid = {0000-0001-9431-6679},
  month = {1},
  title = {{ProphitBet - An Open Source Soccer Prediction App}},
  url = {https://github.com/kochlisGit/ProphitBet-Soccer-Bets-Predictor},
  version = {2.0.0},
  year = {2024}
}