An optimized back propagation neural network on small samples spectral data to predict nitrite in water

Area of Science:

• Environmental Science
• Analytical Chemistry
• Data Science

Background:

• Accurate water pollutant detection is crucial for conservation.
• Small sample sizes and unstable models hinder accurate prediction of nitrite levels.

Purpose of the Study:

• To develop a precise method for predicting nitrite concentration in aquatic environments.
• To overcome limitations of small sample sizes and model instability in spectral data analysis.

Main Methods:

• Dimensionality reduction of spectral data using Kernel Principal Component Analysis (KPCA).
• Sample augmentation via Generative Adversarial Network (GAN) to increase data scale and diversity.
• Optimization of Back Propagation neural network using an enhanced Particle Swarm Optimization (PSO) algorithm for improved training.

Main Results:

• The developed model achieved high prediction accuracy with R² of 0.976, RMSE of 0.0086, and MAE of 0.0066.
• The proposed method demonstrated superior performance compared to state-of-the-art techniques.
• The enhanced PSO algorithm improved model fitting and training performance.

Conclusions:

• The study presents a promising and effective model for accurate nitrite concentration prediction in water.
• The fusion of KPCA, GAN, and optimized PSO with neural networks offers a robust solution for water quality monitoring.