Journal of Basic and Applied Research International

Rising global energy demand alongside environmental constraints has intensified interest in biomass as a sustainable alternative energy source, although its direct utilisation is limited by challenges in thermochemical conversion and fuel quality variability. Torrefaction has therefore emerged as a promising pre-treatment technique for enhancing the energy density and combustion characteristics of biomass, necessitating reliable and cost-effective methods for accurately estimating its heating value for industrial applications. The project aims to design a torrefaction reactor and evaluate the energy content of torrefied coconut shells using artificial neural network (ANNs), random forest and linear regression. The best method for predicting the HHV of coconut shell was determined. The designed reactor enables optimal torrefaction, facilitating the production of high-quality torrefied coconut shell. Machine learning algorithms, including Artificial Neural Networks (ANN), Random Forest, and Linear Regression, are employed to predict the Higher Heating Value (HHV) of the torrefied coconut shell. The evaluation reveals strong correlations between the predicted HHV values and the actual HHV values extracted from literature sources. The ANN model had the highest level of accuracy followed by the linear regression model and then the random forest model. The ANN achieved a Mean Absolute Error (MAE) of 1.399 and Mean Squared Error (MSE) of 4.083 for proximate datasets and a Mean Absolute Error (MAE) of 1.046 and Mean Squared Error (MSE) of 2.565 for ultimate datasets. Torrefied biomass feature importance analysis highlights the significant influence of fixed carbon, ash, and volatile matter on HHV prediction. The findings contribute to understanding the torrefaction process, optimising reactor design, and advancing machine learning techniques for predicting torrefied coconut shell's energy content. The ANN model demonstrated the best predictive performance among the evaluated models, achieving the lowest MAE and MSE values for both proximate and ultimate datasets. The empirical correlations also showed strong agreement with literature HHV values, with ultimate analysis producing slightly better predictive accuracy than proximate analysis. The findings demonstrate the suitability of integrating torrefaction reactor design with machine learning techniques for biomass energy characterization and sustainable waste-to-energy applications.