AIML Project-Based Learning (PBL)

Predictive Analytics for Urban Water Demand

Overview

This project is developed for PCS-253 AIML Project-Based Learning (PBL).

The main objective of this project is to predict and classify urban water consumption using Machine Learning techniques. Multiple models are implemented and compared to improve prediction accuracy and classification performance.

The project is divided into three phases showing progressive improvement in dataset preparation, preprocessing, model training, and evaluation.

---

Project Structure

Phase 1 - Initial Development

Objective

To build a basic machine learning project using the raw dataset.

Work Done

• Created initial dataset
• Basic project implementation
• Problem understanding
• Initial data analysis

Files

• dataset.csv
• report.pdf

---

Phase 2 - Model Improvement

Objective

To improve dataset quality and evaluate machine learning models.

Work Done

• Data cleaning
• Feature analysis
• Data preprocessing
• Model evaluation using:
  • MSE
  • RMSE
  • R² Score

Files

• Phase2.ipynb
• improved_dataset.csv
• report.pdf

---

Phase 3 - Final Implementation

Objective

To build an optimized machine learning system for predicting urban water demand.

Features Used

• Household_Size
• Income_Level
• Garden_Area
• Rainfall_mm
• Temperature_C
• Seasonal_Index
• Water_Price
• Appliance_Count
• Leakage_Factor

Additional Engineered Features

• Per_Person_Usage
• Garden_Impact
• Water_Intensity

Machine Learning Models Used

• Linear Regression
• K-Nearest Neighbors (KNN)
• Support Vector Machine (SVM)
• Decision Tree

Work Done

• Final dataset preparation
• Data preprocessing
• Feature scaling using StandardScaler
• Train-test split
• Model training and prediction
• Model comparison
• Performance evaluation
• Data visualization and graphs

Evaluation Metrics

• Accuracy
• Precision
• Recall
• F1 Score
• R² Score
• Mean Squared Error (MSE)
• Mean Absolute Error (MAE)
• ROC-AUC Curve

Visualizations

• Correlation Heatmap
• Histogram
• Boxplot
• Scatter Plot
• Confusion Matrix
• ROC Curve
• Feature Importance Graph

Best Performing Model

Decision Tree achieved the best performance based on F1 Score.

Files

• Phase3.ipynb
• final_dataset.csv
• report.pdf

---

Technologies Used

• Python
• Pandas
• NumPy
• Matplotlib
• Seaborn
• Scikit-learn
• Jupyter Notebook
• VS Code

---

Workflow

Dataset → Data Cleaning → Feature Engineering → Feature Scaling → Model Training → Prediction → Evaluation → Visualization

---

Conclusion

This project successfully predicts and classifies urban water usage using multiple machine learning models. Different preprocessing and evaluation techniques were applied to improve model performance and compare different algorithms effectively.

---

Future Scope

• Use advanced models like Random Forest and XGBoost
• Real-time water demand prediction
• Deploy using Flask or Streamlit
• Add live dashboard and monitoring system

---

GitHub Repository

Python_PBL Repository