5G User Prediction Project

Project Overview

This project aims to predict whether a user is a 5G user based on their basic information and communication-related data such as call charges, data usage, activity behavior, package type, and region information. The project uses various machine learning algorithms to build predictive models and optimize their performance.

Project Background

During the 2022 World Internet Conference Wuzhen Summit, the "World Internet Development Report 2022" indicated that in the first quarter of 2022, the global number of 5G users increased by 70 million, reaching approximately 620 million. According to Ericsson and GSMA, the global number of 5G users is expected to exceed 1 billion by the end of 2022. For telecom operators, it is helpful to create user profiles based on user-side information and make precise marketing efforts for potential 5G users.

Data Description

The dataset contains 60 fields, with the target field being the prediction target. The main feature fields are divided into two categories: categorical (cat) and numerical (num) features. The dataset is available on the cloud platform, named train.csv.

Feature Fields

• id: Sample identifier
• cat_0 to cat_19: Categorical features
• num_0 to num_37: Numerical features
• target: Target field, indicating whether the user is a 5G user

Evaluation Metrics

The performance of the models is evaluated using the AUC (Area Under the Curve) metric. The higher the AUC score, the better the model performance.

Algorithms and Optimization Techniques

Initial Models

1. Adaboosting
2. Decision Tree
3. K-means Clustering
4. KNN with FAISS
5. LightGBM
6. Logistic Regression
7. Naive Bayes
8. Random Forest
9. XGBoost

Optimization Techniques

1. Grid Search: Used in KNN_Grid Search.ipynb, LightGBM_Grid Search.ipynb, and XGBoost_Hyperparameter Search.ipynb.
2. K Neighbors Change: Used in KNN_K_change.ipynb, KNN_K_n_neighbors_change.ipynb, and KNN_n_neighbors_change.ipynb.
3. Bayesian Optimization: Used in LightGBM_Bayes.ipynb and XGBoost_Bayes_3para.ipynb.
4. Particle Swarm Optimization (PSO): Used in LightGBM_PSO.ipynb.
5. Feature Crosses: Used in XGBoost_Feature Crosses.ipynb.
6. Random Search with Early Stopping: Used in XGBoost_Random Search_Early Stopping.ipynb.

Repository Structure

Content

5Gpredict_Adaboosting.ipynb 5Gpredict_DecisionTree.ipynb 5Gpredict_K-means.ipynb 5Gpredict_KNN_FAISS.ipynb 5Gpredict_LightGBM.ipynb 5Gpredict_LogisticRegression.ipynb 5Gpredict_NaiveBayes.ipynb 5Gpredict_Randomforest.ipynb 5Gpredict_XGBoost.ipynb KNN_Grid Search.ipynb

KNN

1. KNN_K_change.ipynb
2. KNN_K_n_neighbors_change.ipynb
3. KNN_n_neighbors_change.ipynb

LightGBM

1. LightGBM_Bayes.ipynb
2. LightGBM_Grid Search.ipynb
3. LightGBM_PSO.ipynb

XGBoost

1. XGBoost_Feature Crosses.ipynb
2. XGBoost_Hyperparameter Search.ipynb
3. XGBoost_Bayes_3para.ipynb
4. XGBoost_Random Search_Early Stopping.ipynb

data

train.csv (not included in this repository)

How to Run

1. Clone the repository:

   git clone https://github.com/F4nc1est/5GUserPrediction.git

2. Navigate to the project directory:

   cd 5GUserPrediction

3. Run the Jupyter notebooks to see the modeling process and results:

   jupyter notebook

Results and Analysis

The detailed results and analysis of each model, including performance comparisons and optimization effects, can be found in the respective Jupyter notebooks. The analysis includes the reasoning behind model selection, data analysis, model comparison, and potential improvement suggestions.

Contributing

We welcome contributions to this project. Please create a pull request or open an issue to discuss any changes.

License

This project is licensed under the MIT License. See the LICENSE file for details.