arnaudbozahbe-afk/Prediction-model-for-the-Conservation-of-Annonaceae-Species-in-Central-Africa

Artificial Intelligence (AI) paradigms for predicting the conservation status of Annonaceae species in Central Africa # Abstract The accelerating biodiversity crisis—driven by climate change and intensified human activities—demands faster, data-driven, and more systematic approaches to assess species extinction risk beyond traditional IUCN Red List protocols. This study tackles this challenge by evaluating and comparing **two Artificial Intelligence (AI) paradigms** for predicting the conservation status of *Annonaceae* species in Central Africa: --- ## 1. Supervised Classification (Machine Learning / Deep Learning) This paradigm models the direct relationship between species characteristics and their IUCN threat category. It leverages a diverse set of predictors to classify species as threatened or non-threatened. --- ## 2. Link Prediction in Knowledge Graphs (KGs) This paradigm explores symbolic AI by: - constructing a knowledge graph integrating biological, ecological, and phylogenetic data - inferring missing conservation statuses using **link prediction** and graph-based reasoning --- ## Dataset & Innovation A major contribution of this study lies in the integration of a **heterogeneous, multi-source dataset** including: - **2,500 species** - **155 variables**, covering: - morphological traits - ecological metrics (EOO / AOO) - bioclimatic variables - phylogenetic information - IUCN conservation data - **Near-Infrared Spectroscopy (NIR) data**, included for the first time in this context Extensive preprocessing was employed, including **hierarchical clustering** to identify ecological subgroups prior to model training. --- ## Key Results ### ✔ Supervised Learning The **Support Vector Machine (SVM)** model achieved the strongest performance: - **Accuracy:** 80% - **F1-score:** 0.74 - Best performance recorded for species in **Cluster 1** ### ✔ Knowledge Graph Approach The knowledge-graph framework demonstrated the feasibility of **symbolic inference** for conservation assessment, offering a complementary perspective to purely numeric ML models. --- ## Concl …