[Paper Review] Predicting Pollution Level Using Random Forest: A Case Study of Marilao River in Bulacan Province, Philippines

Jayson M. Victoriano, Manuel Luis C. Delos Santos+2 more|RePEc: Research Papers in Economics|Feb 12, 2022

Water Quality Monitoring Technologies4 references4 citations

TL;DR

This study develops a Random Forest-based machine learning model to predict pollution levels in the Marilao River, Philippines, using water quality parameters (DO, pH, BOD, TSS). The model achieved 91.75% accuracy and Kappa value of 0.8115, indicating strong predictive performance for classifying pollution as Green (Normal), Yellow (Average), Orange (Polluted), or Red (Highly Polluted).

ABSTRACT

This study aims to predict the pollution level that threatens the Marilao River, located in the province of Bulacan, Philippines. The inhabitants of this area are now being exposed to pollution. Contamination of this waterway comes from both formal and informal industries, such as a used lead-acid battery, open dumpsites metal refining, and other toxic metals. Using various water quality parameters like Dissolved Oxygen (DO), Potential of Hydrogen (pH), Biochemical Oxygen Demand (BOD) and Total Suspended Solids (TSS) were the basis for predicting the pollution level. This study used the Data Mining technique based on the sample data collected from January of 2013 to November of 2017. These were used as a training data and test results to predict the river condition with its corresponding pollution level classification indicated with the used of colors such as Green for Normal, Yellow for Average, Orange for Polluted and Red for Highly Polluted. The model got an accuracy of 91.75% with a Kappa value of 0.8115, interpreted as Strong in terms of the level of agreement.

Motivation & Objective

• To address growing pollution concerns in the Marilao River, a critical water source in Bulacan Province, Philippines.
• To analyze the impact of formal and informal industrial activities, including lead-acid battery recycling and open dumpsites, on river water quality.
• To develop a predictive model for classifying pollution levels using machine learning on historical water quality data.
• To provide a color-coded pollution classification system (Green, Yellow, Orange, Red) for actionable environmental monitoring.
• To evaluate the performance of the Random Forest algorithm in predicting pollution levels with high accuracy and strong agreement metrics.

Proposed method

• Collected and analyzed water quality data from January 2013 to November 2017 for key parameters: Dissolved Oxygen (DO), pH, Biochemical Oxygen Demand (BOD), and Total Suspended Solids (TSS).
• Applied Data Mining techniques to preprocess and structure the dataset for machine learning modeling.
• Trained a Random Forest classifier using the historical dataset to predict pollution levels based on input water quality parameters.
• Classified pollution levels into four categories: Green (Normal), Yellow (Average), Orange (Polluted), and Red (Highly Polluted), using color-coded labels.
• Evaluated model performance using accuracy and Cohen’s Kappa coefficient to assess inter-rater agreement and predictive reliability.
• Used a 10-fold cross-validation approach to ensure robustness and generalization of the model across the dataset.

Experimental results

Research questions

• RQ1 Can a Random Forest model accurately predict pollution levels in the Marilao River using standard water quality parameters?
• RQ2 How effective is the Random Forest algorithm in classifying pollution levels into distinct categories (Green, Yellow, Orange, Red) based on DO, pH, BOD, and TSS?
• RQ3 What is the level of agreement between predicted and actual pollution classifications, as measured by the Kappa statistic?
• RQ4 To what extent can machine learning models support environmental monitoring and early warning systems for river pollution in industrialized regions?
• RQ5 How do the selected water quality parameters (DO, pH, BOD, TSS) contribute to the predictive power of the model?

Key findings

• The Random Forest model achieved a prediction accuracy of 91.75% on the test dataset, indicating strong predictive capability.
• The model's Cohen’s Kappa value was 0.8115, interpreted as 'Strong' agreement between predicted and actual pollution classifications.
• The model successfully classified pollution levels into four distinct categories: Green (Normal), Yellow (Average), Orange (Polluted), and Red (Highly Polluted).
• The inclusion of DO, pH, BOD, and TSS as input features significantly contributed to the model’s predictive performance.
• The model demonstrated robustness and generalization through 10-fold cross-validation, ensuring consistent results across data splits.
• The study confirms that machine learning, particularly Random Forest, is a viable and effective tool for environmental pollution prediction in data-scarce, industrial-impacted river systems.