MACHINE LEARNING MODELS FOR PREDICTING STRESS VALUE IN THE TENSILE STRENGTH OF BIOFILMS FROM STARCH AND HAIR WASTE

• Published in: Jurnal pena sains Vol. 11; no. 2; pp. 54 - 62
• Main Authors: Utami, Hermin Hardyanti, Fitrah, Muhammad Aqdar, Yusriadi, Yusriadi, Ardiansah, Ardiansah, Arminas, Arminas, Lestari, Mega Fia, Towolioe, Sherly
• Format: Journal Article
• Language: English
• Published: 31.10.2024
• ISSN: 2407-2311; 2527-7634
• DOI: 10.21107/jps.v11i2.26227

Biofilms, structured communities of microorganisms, have emerged as a subject of significant interest across various industries due to their unique biodegradable and sustainable characteristics. Hair waste is an incredibly rich source of keratin, and this abundance makes it a promising candidate as a fundamental building block for the development of biodegradable plastics. This study focuses on sustainable biofilms derived from biodegradable materials, specifically a unique combination of starch and hair waste. Machine Learning models, implemented in RapidMiner, were utilized to predict the tensile strength of these biofilms, with the goal of enhancing quality control in their production. Neural Networks and Deep Learning methods were employed to compare their predictive capabilities, assessing both their strengths and limitations. Through rigorous data collection, feature identification, and detailed data analysis, critical factors influencing the quality of the biofilms were identified. The results revealed the remarkable predictive accuracy of the Neural Net model, particularly for Ratio 40, while the performance of the Deep Learning model varied across different ratios. The lower RMSE of the Neural Net model indicated a more precise alignment between the predicted and actual values, distinguishing it as the superior model. This research contributes to the advancement of sustainable biofilm development, offering eco-friendly solutions through the use of unconventional materials. Both models offer valuable predictive capabilities, and the choice between them may depend on the specific requirements and contexts of the application. In conclusion, the performance of the Neural Net and Deep Learning models in predicting stress in tensile strength varies across different ratios.