Browsing by Author "Maina, Thuku Jeremiah"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Rapid Detection of Chlorpyrifos in Kale and Milk Using Machine Learning-Aided Raman Spectroscopy
    (Kenyatta University, 2025-12) Maina, Thuku Jeremiah
    Chlorpyrifos, a widely used organophosphorus pesticide in Kenya, is banned from use on vegetables due to its health risks; however, studies show it is still widely used and detected in food products. Conventional detection methods, such as gas chromatography (GC) and high-performance liquid chromatography (HPLC), are accurate but costly, time-consuming, and destructive, making them unsuitable for rapid on-site analysis. This study aimed to develop a fast, non-destructive method for detecting chlorpyrifos in milk and kale using Raman spectroscopy and machine learning (ML). ML involves computational algorithms that analyze complex data patterns, improving prediction accuracy and classification. These techniques were crucial for efficiently processing spectral data, recognizing patterns, and building predictive models for chlorpyrifos detection. Raman spectroscopy was chosen for its solvent-free, non-invasive nature. Spectral preprocessing steps, including baseline correction, smoothing, and normalization, improved signal quality. Analysis of Variance (ANOVA) was applied to identify Raman bands with statistically significant differences, and Principal Component Analysis (PCA) revealed the spectral fingerprint and reduced dimensionality. The 314-354 cm⁻¹ spectral band, centered at 342 cm⁻¹, was identified as the chlorpyrifos Raman fingerprint due to distinct C-Cl vibrational modes absent in untreated samples. Machine learning models, including Support Vector Machine (SVM), Support Vector Regression (SVR), and Random Forest (RF), were trained using Principal Components (PCs) from the fingerprint. These models were used to classify chlorpyrifos levels in the samples with respect to the Maximum Residue Limit (MRL), the highest permissible pesticide concentration in food for consumer safety, ensuring the models provided relevant food safety assessments. Classification models achieved high accuracy: SVM outperformed RF with 95.79% accuracy in milk and 92.61% in kale, while RF achieved 95.23% and 90.15%, respectively. In regression tasks, RF showed superior performance with a coefficient of determination (R²) > 0.9997 and a root mean square of prediction (RMSEP) < 0.0231 ppm, compared to SVR’s R² > 0.9961 and RMSEP < 0.0897 ppm. These results confirm that Raman spectroscopy combined with ML offers a highly accurate, rapid, and non-destructive alternative to conventional methods, enhancing real-time food safety monitoring and regulatory compliance