Advanced Agricultural Decision System Using Recurrent Polynomial Network for Multi-Crop Recommendation Tasks

Abstract

Agriculture has significantly transitioned from traditional, experience-based practices to modern datadriven approaches with the integration of digital technologies and Machine Learning (ML). Earlier, crop monitoring and decision-making relied heavily on manual observation, farmer expertise, and basic statistical techniques. However, the rapid growth of agricultural data from sensors, weather records, and satellite imagery has created a need for intelligent systems capable of handling large and complex datasets efficiently. Traditional methods often fail to capture intricate relationships among variables such as soil characteristics, climate conditions, and crop health, leading to lower prediction accuracy, delayed decisions, and inefficient resource usage. This study addresses the limitation of existing systems that perform classification and regression tasks separately, resulting in fragmented and inconsistent outputs. To overcome this, a unified multi-task agricultural framework based on the 1CA2RT (One Classification and Two Regression Tasks) approach is proposed. The system performs crop disease classification along with NDVI estimation and harvest time prediction simultaneously. The framework employs ML models including Support Vector Machine (SVM)-1CA2RT, AdaBoost (AB)- 1CA2RT, and Ridge (R)-1CA2RT, along with a Deep Learning-based Hybrid Recurrent Polynomial Ensemble (HRPE-1CA2RT) model. This model integrates a Bidirectional Long Short-Term Memory (Bidirectional LSTM)-based Recurrent Polynomial Network with ensemble Tao Tree methods to capture complex nonlinear patterns. The system is implemented using Flask and SQLite for secure, real-time interaction. Experimental results demonstrate outstanding performance, achieving perfect classification and regression accuracy.