검색 상세

Water Quality Prediction for Caranx ignobilis Aquaculture Pond in the Philippines Using Machine Learning Models

  • MAE ANNE P. GARDON (한동대학교 국제개발협력대학원 국제개발협력학과)

초록/요약

Aquaculture is a key contributor to global food security, with fish farming providing a substantial portion of fish production. In the Philippines, Caranx ignobilis (Giant Trevally) is a highly valued species in freshwater aquaculture. However, the sensitivity of C. ignobilis to water quality fluctuations, particularly dissolved oxygen (DO) levels, poses a significant challenge, often leading to fish mortality and economic losses. This study aims to address these challenges by applying machine learning (ML) techniques to predict DO levels in aquaculture ponds, helping aquaculture managers maintain optimal water conditions. The study employed three machine learning models: Long Short-Term Memory (LSTM) networks, Support Vector Regression (SVR), and Random Forest Regression (RFR). Dataset used for the this study contains 10 features and 2,328 readings of water quality parameters collected from the Freshwater Fisheries Research and Development Center (FFRDC) over a four and a half year period. Additionally, atmospheric data were incorporated to enhance the predictive power of the models. The dataset was split, with 80% used for model training and 20% for testing. Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percent Error (MAPE), and R-squared (R²) are the evaluation metrics used to assess model performance. The LSTM model achieved the best results, with an MAE of 0.86, RMSE of 1.13, MAPE of 3%, and an R² of 0.77, effectively capturing temporal dependencies in the data. This research highlights the potential of ML models in enhancing aquaculture operations, contributing to the sustainability and resilience of the industry.

more

목차

I. Introduction 1
1. Problem Statement 6
2. Objectives and Goals 9
3. Research Questions 10
4. Significance of the Study 11
II. Literature Review 13
1. Caranx ignobilis Aquaculture in the Philippines 13
2. Water Quality Management in Aquaculture 17
3. Machine Learning (ML) 19
4. Regression-Based ML Models for Timeseries Prediction 21
4.1 Linear Regression 21
4.2 Decision Tree and Random Forest 22
4.3 Support Vector Machine (SVM) 22
4.4 Artificial Neural Networks (ANN) and Long Short-Term Memory (LSTM) Network 22
4.5 Gradient Boosting Machine (GBM) 23
5. ML Application in Aquaculture Water Quality Management 23
6. ML Models for DO Prediction in Aquaculture 24
III. Materials and Methods 27
1. Data Collection 29
1.1 Dataset Overview 33
2. Data Pre-processing 34
2.1 Data Loading and Inspection 34
2.1 Handling Missing Values 35
2.3 Conversion of Data Types and Interpolation 35
3. Data Analysis 36
3.1 Descriptive Statistics 36
3.2 Data Visualization 37
3.3 Correlation Analysis 38
4. Model Development 39
4.1 Model Selection Rationale 39
4.1.1 Support Vector Regression (SVR) 47
4.1.2 Random Forest Regression (RFR) 50
4.1.3 Long Short Term Memory (LSTM) Network 53
4.2 Model Training 56
4.2.1 Normalization/Scaling 56
4.2.2 Feature Engineering 57
4.2.3 Hyperparameter Tuning 59
4.2.4 Model Parameters 60
4.3 Model Testing 66
5. Model Evaluation 67
5.1 Mean Absolute Error (MAE) 67
5.2 Root Mean Square Error (RMSE) 68
5.3 Mean Absolute Percent Error (MAPE) 69
5.4 R-squared (R2) 69
IV. Results 71
1. Exploratory Data Analysis Result 71
1.1 Descriptive Statistics 71
1.2 Data Visualization 72
1.2.1 Time-Series Plot 72
1.2.2 Data Distribution 73
1.2.3 Pair Plot 76
1.2.4 Grouping Analysis 77
1.3 Correlation Analysis 79
2. Model Performance 83
2.1 DO Prediction Using SVR Model 84
2.2 DO Prediction Using RFR Model 87
2.3 Time-Series Prediction Using LSTM Model 90
V. Discussion 93
VI. Conclusion 100
VII. Recommendation 104
VIII. References 107
Appendix 1 – Policy Plan 137

more
반출 Meta View 목록