Fusion-based machine learning approach for classification of algae varieties exposed to different light sources in the growth stage

Gerdan Koc D., KOÇ C., Ekinci K.

Algal Research, cilt.71, 2023 (SCI-Expanded) identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 71
  • Basım Tarihi: 2023
  • Doi Numarası: 10.1016/j.algal.2023.103087
  • Dergi Adı: Algal Research
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aquatic Science & Fisheries Abstracts (ASFA), BIOSIS, Food Science & Technology Abstracts, INSPEC
  • Anahtar Kelimeler: Classification, LED light, Machine learning, Microalgae
  • Ankara Üniversitesi Adresli: Evet

Özet

Microalgae are one of the most important organisms in the ecosystem, and they have important roles in terms of their function and continuity in the ecosystem. Additionally, microalgae have come to the fore as an alternative energy source with both clean energy and high efficiency in recent years. Machine learning approach has become important and effective in recent years due to its wide-ranging applications. In this study, three different microalgae species, which are Chlorella kessleri (UTEX 398), Botryococcus braunii (UTEX 572), and Synechococcus leopoliensis (UTEX B 625), were cultivated in a controlled environment using three different light sources (red LED, blue LED, and fluorescent). The data obtained at the end of the study were used to estimate the classification according to the cultivation parameters (cell count, biomass weight, pH, ORP (oxidation-reduction potential), temperature, CO2 concentration in air, and dissolved oxygen) that are important in microalga cultivation. The algal growth dataset was partitioned using the 10-fold-cross validation method. The Random Forest (RF), Gradient Boosted Trees (GBT), and Naïve Bayes (NB) were selected as machine learning algorithms. Furthermore, the RF-GBT fusion was computed to improve the accuracy rate. The best prediction was found to be 93.11 % for the RF-GBT fusion-based machine learning algorithm. The accuracy of RF, GBT, and NB was determined as 93.06 %, 90.28 %, and 86.11 %, respectively.