A Novel MLDA-MRF Framework for Crop Yield Prediction Modeling with Feature Reduction

Abstract

Agricultural data analysis poses unique challenges due to the multidimensional nature of datasets and the complex interactions between various factors affecting crop yield and soil health. In this research work, we present a innovative method to focus these confronts by combining Modified Linear Discriminant Analysis (MLDA) for feature reduction with Modified Random Forest (MRF) for prediction modeling. We utilize a dataset sourced from the Indian Chamber of Food and Agriculture (ICFA), focusing on essential soil parameters including nitrogen (N), phosphorus (P), potassium (K), and soil pH values. The first phase of our methodology involves data preprocessing to ensure data cleanliness and normalization. We then employ MLDA, tailored specifically for agricultural data, to identify the most discriminative features among the dataset. By incorporating domain-specific knowledge, MLDA effectively selects the key variables influencing agricultural outcomes, such as crop yield and soil fertility. Subsequently, we utilize MRF, a robust ensemble learning algorithm, to build predictive models based on the reduced feature set obtained from MLDA. MRF is chosen for its capability to treat higher-dimensional data and provide accurate predictions, crucial for decision-making in agriculture. Through extensive experimentation and evaluation, we assess the performance of the MLDA-MRF framework in terms of R2 score, MAE, RMSE and accuracy. Our results demonstrate the efficacy of the proposed approach in both feature reduction and prediction tasks, outperforming traditional methods. This research contributes to advancing agricultural data analysis by providing insights into the significant factors influencing agricultural parameters. The proposed methodology not only aids in optimizing agricultural practices but also facilitates informed decision-making, thereby contributing to sustainable agriculture and food security. The integration of MLDA and MRF offers a promising avenue for analyzing agricultural datasets, enabling stakeholders to make data-driven decisions for improved productivity and resource management in the agricultural sector.