Page 12 - AJWEP-22-4

Ujianti, et al.

quality measurement. The weighted arithmetic water allow for early detection of pollution events, real-
quality index is employed alongside machine learning time diagnostics, and longitudinal analysis through
models, specifically random forest, LightGBM, and historical trend monitoring – capabilities often absent
XGBoost, to predict water quality. Interpretation of in comparable low-cost IoT implementations. The
model predictions using Shapley additive explanations proposed system offers a uniquely practical solution for
reveals that chemical oxygen demand and biological continuous water quality monitoring in low-resource,
oxygen demand (BOD) are the most influential factors high-risk coastal areas by combining high-frequency
in determining water quality. Meanwhile, electrical data capture, environmental adaptability, affordability,
conductivity (EC), chloride, and nitrate have minimal and local relevance.
impact. Machine learning and IoT can be integrated
36
with one comprehensive water quality monitoring 2. Materials and methods
framework that includes sensor architecture and
communication protocols to address field challenges. 2.1. Materials
IoT sensors reduce contamination detection time, Water sampling was conducted at Banjardowo River,
enable early warning and rapid response compared Semarang city, and Buntu River, Kendal Regency in
to conventional laboratory methods, and provide a Indonesia. Parameters of the water samples such as
comprehensive synergy of machine learning-IoT, temperature, pH, and TDS were tested, considering that
besides focusing on XAI and ethical access issues. these parameters describe river water quality and require
The advantages are high performance, practical impact direct testing. This IoT testing was carried out at the
for early detection, global policy relevance, and clear Laboratory of Mechanical Engineering in Engineering
research directions. 37 and Informatics Laboratory of Universitas PGRI
In response to these limitations, IoT technology has Semarang, Semarang city, Central Java, Indonesia.
emerged as a promising alternative, offering continuous,
automated, and real-time data collection. IoT-based 2.2. Design of water quality monitoring system
water monitoring systems leverage sensor networks and A microcontroller called NodeMCU was used to
microcontrollers to measure water quality parameters incorporate IoT into water quality testing. NodeMCU is
and transmit data to cloud-based platforms, facilitating an open-source IoT platform and a development kit that
instant access, analysis, and alerting. Prior research uses programming language to help create prototypes
has implemented IoT in aquaculture management, of IoT products. It is compatible with sketches created
wastewater control, and agricultural runoff monitoring with Arduino. During this test, the water quality levels
38
applications. However, many systems are constrained from several rivers in the coastal areas were studied.
by limited parameter coverage, low temporal resolution, The stages in the design of this water quality monitoring
lack of historical data integration or insufficient system include (i) system design, (ii) component
calibration, and field validation. assembly, (iii) system testing, (iv) system trial, and
The research gap in the study and design of this tool is (v) validation.
compared with the previous research mentioned above The electronic architecture of the proposed IoT-based
because it can monitor water quality at a per-second water quality monitoring system was centered on the
frequency. This study presents an IoT-based water ESP32 microcontroller, which served as the primary
quality monitoring system that differs from existing processing unit integrating multiple sensor inputs and
approaches in several critical ways. Unlike most peripheral modules. The ESP32 interfaced with an array
systems that sample at long intervals (minutes or hours), of water quality sensors – including pH, turbidity, TDS,
this system acquires data at a per-second frequency, temperature, and color sensors – through its analog-
capturing rapid fluctuations that are often missed. It to-digital converter channels, enabling precise and
integrates four key sensors – pH, TDS, temperature, real-time data acquisition (DAQ) of the environmental
and turbidity – into a compact, N ode Microcontroller parameters. A Nextion liquid crystal display (LCD)
Unit (NodeMCU), which includes onboard data display, module was connected through a universal asynchronous
cloud data logging, and sensor calibration protocols. receiver/transmitter interface, providing an intuitive
Furthermore, the system is validated against laboratory- graphical user interface for immediate visualization of
grade instruments and tested in actual coastal field sensor data and system status. Data logging capabilities
environments, demonstrating technical performance were facilitated by a secure digital card module
and environmental robustness. These innovations interfaced through the serial peripheral interface

Volume 22 Issue 4 (2025) 4 doi: 10.36922/AJWEP025110069