Authors: Raquel Cossío-Ramírez, Hortencia A. Ramírez-Vázquez, Amanda M. Valdez-Calderón
Mexico faces significant challenges in water contamination due to various factors, including natural contaminants, industrial pollution, agricultural runoff, urbanization, and overexploitation. As a result, 57% of the population lacks access to safe water.
This study’s objective is to create a semi-supervised ensemble model capable of predicting water quality by municipality based on environmental factors, such as population, inundability and industrial activity.
The project focuses on extending water quality classification coverage across Mexico by combining a small set of labeled data with a large amount of unlabeled data. A self-training method using ensemble classifiers (e.g., Random Forest) is implemented to label previously unclassified municipalities.
-
Data Preprocessing Data is cleaned and prepared in
Data cleaning1.ipynb. -
Model Training Semi-supervised learning is implemented in
self_learning.ipynbusing a self-learning wrapper with ensemble classifiers. -
Evaluation & Visualization Model performance is assessed using confusion matrices, feature importance, and label maps.
In order to predict water quality for municipalities that were not covered by measuring stations, we opt to build a dataset with features that could be indirectly related to water quality. We included features per municipality such as population, altitude, area, flood-ability threshold, number of water treatment facilities and wastewater discharge points, and number of people working in industries. In this notebook AllData.csv is the integration of the datasets metioned above.
- File:
AllData.csv - Level: Municipality-level indicators across Mexico
- Includes: Both labeled (classified) and unlabeled samples
Input dataset: Flood-ability threshold according to precipitation and land permeability
Input dataset: Altitude per locality
Input dataset: Geospatial information on municipal management of water and sanitation services
Input dataset: Human Impact Index on Terrestrial Biodiversity
Input dataset: Economic activity. Agriculture, animal husbandry and farming, forestry, fishing, and hunting
Input dataset: Economic activity. Construction