Abstract:
Effective management of Haplodiploid spider mites (Tetranychus ludeni) is a significant agricultural
challenge due to their complex, non-linear population dynamics, which are poorly handled
by traditional Integrated Pest Management (IPM) strategies based on static thresholds. This research
presents the development and evaluation of an integrated AI-driven system for proactive
pest management. The system architecture combines three core modules: (1) a computer vision
module, using a lightweight YOLOv8 Nano model, for automated mite detection; (2) a forecasting
module to predict population trends using machine learning and deep learning; and (3) a
web-based advisory system to provide agricultural instructors with actionable insights based on
a static Economic Injury Level (EIL) derived from observational data. This research documents
the initial exploratory data analysis, synthetic data generation, and a comparative evaluation of
several predictive models. The claim for the tuned XGBoost model lies in a very good performance
with a value of R2 = 0.8351 for the intermediate count prediction. Long Short-Term
Memory (LSTM) and Autoregressive Integrated Moving Average (ARIMA) time-series models
have been tested to forecast population counts over time. The results explicitly show that,
though there is potential in using machine learning for such forecasting tasks, there are considerable
challenges involved in applying time-series models that need further refinement. The
time series features introduced XGBoost model lies in a very good performance with a value of
R2 = 0.9995 for the future count prediction (trained on synthetic data tested on real data). However,
this project has amalgamated these components toward building and validating a coherent
tool driving enhanced agricultural sustainability practice through minimized crop loss arising
from optimised pest control intervention.
