Abstract
Open-access sensor datasets are increasingly used to train Machine Learning (ML) models for aquaponics monitoring, yet their fitness for purpose is rarely audited. We benchmark five ML algorithms for turbidity prediction on a widely used dataset under three evaluation protocols: a standard random train-test split, a temporal (chronological) split, and a structured forecasting pipeline with short-horizon targets; we further apply feature-importance analysis to identify dominant predictors. All five algorithms achieve $R^{2} ≥ 0.95$ under a random split but collapse to $R^{2}$ between -5.28 and -14.06 under a chronological split, a gap driven by sensor saturation (31.6% of readings), physically impossible negative values (9.9%), and severe temporal distribution shift. The forecasting pipeline partially recovers short-horizon performance (best $R^{2}=0.90$ at 15min and $R^{2}=0.87$ at 1 h), but data quality imposes a hard ceiling on robust chronological generalisation. Feature-importance analysis consistently ranks pH as the dominant predictor (35-43% across RF Gini and XGBoost gain), with temperature contributing about 15%, supporting cost-aware sensor selection for resource-limited deployments.
| Original language | English |
|---|---|
| Title of host publication | Proceedings - 2026 IEEE 50th Annual Computers, Software, and Applications Conference, COMPSAC 2026 |
| Editors | Hossain Shahriar, Kazi Shafiul Alam, Hiroyuki Ohsaki, Stelvio Cimato, Miriam Capretz, Shamem Ahmed, Sheikh Iqbal Ahamed, AKM Jahangir Alam Majumder, Munirul Haque, Tomoki Yoshihisa, Alfredo Cuzzocrea, Michiharu Takemoto, Nazmus Sakib, Marwa Elsayed |
| Publisher | Institute of Electrical and Electronics Engineers Inc. |
| Pages | 2805-2810 |
| Number of pages | 6 |
| ISBN (Electronic) | 9798331544973 |
| DOIs | |
| Publication status | Published - 10 Jul 2026 |
| Event | 50th IEEE Annual Computers, Software, and Applications Conference, COMPSAC 2026 - Madrid, Spain Duration: 7 Jul 2026 → 10 Jul 2026 |
Publication series
| Name | Proceedings - 2026 IEEE 50th Annual Computers, Software, and Applications Conference, COMPSAC 2026 |
|---|
Conference
| Conference | 50th IEEE Annual Computers, Software, and Applications Conference, COMPSAC 2026 |
|---|---|
| Country/Territory | Spain |
| City | Madrid |
| Period | 7/07/26 → 10/07/26 |
Bibliographical note
Publisher Copyright:© 2026 IEEE.
Keywords
- agricultural IoT
- aquaponics
- machine learning
- sensor data quality
- turbidity forecasting
Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver