charliesix6/water-pump-classification

Clasificación multiclase del estado de bombas de agua en Tanzania. EDA, preprocesamiento y modelos de ML con Python y scikit-learn. # Predictive Maintenance of Water Pumps in Tanzania ## Project Overview This project focuses on predicting the operational status of water pumps in Tanzania using Machine Learning techniques. The objective is to classify pumps into three categories: - Functional - Functional but needs repair - Non-functional The project was developed as an end-to-end Data Analytics & Machine Learning workflow, covering: - Exploratory Data Analysis (EDA) - Data Cleaning & Feature Engineering - Handling Missing Values - Encoding Categorical Variables - Class Imbalance Treatment - Model Training & Evaluation - Kaggle-style Competition Submission --- # Business Problem Access to clean water is a critical issue in many regions of Tanzania. Predicting which water pumps are likely to fail can help organizations prioritize maintenance efforts and improve resource allocation. Using historical operational data, this project builds predictive models capable of identifying malfunctioning pumps before complete failure. # Tech Stack - Python - Pandas - NumPy - Scikit-learn - XGBoost - Imbalanced-learn (SMOTE) - Matplotlib / Seaborn - Jupyter Notebook # Dataset The dataset contains operational and geographical information about water pumps across Tanzania. ### Main Features - GPS coordinates - Population served - Water quantity - Pump type - Installer - Management type - Construction year - Extraction type - Geographic region Target variable: - `status_group` # Exploratory Data Analysis During the initial analysis: - Numerical and categorical variables were identified - Missing values and anomalous values were detected - High-cardinality categorical features were analyzed - Strong class imbalance was identified in the target variable ### Key Findings - Several columns contained a high percentage of missing values - Some numerical columns contained impossible values (e.g. longitude = 0) - Multiple categorical variables had extremely high cardinality - The target classes were …