⚡ Household Power Consumption & Telco Churn Prediction Projects

This repository contains two end-to-end Machine Learning applications:

1. Household Power Consumption Prediction (Streamlit App)
2. Telco Customer Churn Prediction (Tkinter GUI App)

Both projects demonstrate full ML pipelines including preprocessing, training, evaluation, and real-time prediction interfaces.

---

📊 1. Household Power Consumption — Streamlit ML App

📌 Overview

This Streamlit application trains a Random Forest Regressor on the Household Power Consumption dataset and provides:

• Model training & evaluation
• Interactive visualizations
• Real-time prediction interface

---

🚀 Features

• 📂 Upload dataset (.txt or .csv)

• 🧹 Data preprocessing:

  • Combines Date + Time
  • Handles missing values

• 🌲 Model: RandomForestRegressor

• 📈 Model evaluation:

  • Mean Squared Error (MSE)
  • Mean Absolute Error (MAE)
  • R² Score

• 📊 Visualizations:

  • Distribution of Global_active_power
  • Actual vs Predicted scatter plot

• 🔮 Real-time prediction form via Streamlit

---

🛠️ Installation

pip install streamlit pandas numpy matplotlib scikit-learn

---

▶️ Run the App

streamlit run app.py

---

📂 Dataset Requirements

• File: household_power_consumption.txt
• Separator: ;
• Missing values: ?

Required Columns:

• Date
• Time
• Global_active_power (Target)
• Global_reactive_power
• Voltage
• Global_intensity
• Sub_metering_1
• Sub_metering_2
• Sub_metering_3

---

⚙️ Model Details

• Train/Test Split: 80/20
• Model: RandomForestRegressor
• n_estimators: 30
• max_depth: 10
• Scaling: StandardScaler

---

📌 Suggestions for Improvement

• Use SimpleImputer instead of dropping NaNs
• Add feature engineering (hour, day, month)
• Save model using joblib
• Add caching with Streamlit
• Hyperparameter tuning (GridSearchCV)

---

📞 2. Telco Customer Churn Predictor (Tkinter GUI)

📌 Overview

This desktop application predicts whether a telecom customer will churn using a LightGBM classifier with a simple Tkinter GUI.

It also handles class imbalance using SMOTE.

---

🚀 Features

• 📂 Load CSV dataset and train model instantly

• ⚡ LightGBM classifier (high performance)

• ⚖️ SMOTE for imbalance handling

• 📊 Evaluation metrics:

  • Accuracy
  • ROC-AUC Score
  • Precision
  • Recall
  • F1-score

• 🔮 Real-time prediction using GUI inputs

---

🛠️ Installation

pip install pandas scikit-learn imbalanced-learn lightgbm

---

▶️ Run the App

python churn_gui.py

---

📂 Dataset Requirements

Required Columns:

• customerID
• gender
• SeniorCitizen
• Partner
• Dependents
• tenure
• PhoneService
• MultipleLines
• InternetService
• Contract
• PaymentMethod
• MonthlyCharges
• TotalCharges
• Churn (Target)

---

🔮 Prediction Inputs

Only 4 inputs required:

• tenure
• MonthlyCharges
• TotalCharges
• SeniorCitizen (0 = No, 1 = Yes)

---

🧪 Example

Input:

tenure = 2
MonthlyCharges = 85
TotalCharges = 170
SeniorCitizen = 1

Output:

Prediction: Churn
Probability: 0.67

---

📌 Notes & Improvements

• Use encoding pipelines for categorical features
• Add model persistence (joblib)
• Improve GUI layout (frames, validation)
• Try additional models (XGBoost, CatBoost)
• Add feature importance visualization

---

⭐ Author

Built as part of supervised machine learning practice projects.