A CNN-LSTM-Attention-Based Decision Support Model for Land Use and Agricultural Investment Optimization

Abstract

The rapid growth of the global population and urbanization has made the efficient utilization of land resources and agricultural investment decision-making critical for sustainable agricultural development. This study focuses on the application of machine learning, an advanced artificial intelligence technology, for land use optimization and agricultural investment decision-making, where its data analysis and prediction capabilities offer significant potential for improving decision-making processes. Based on machine learning algorithms, this paper studies the construction of land use and agricultural investment decision-making models, aiming at improving the allocation efficiency of agricultural resources through data-driven methods and providing scientific basis for agricultural investment decisions. This paper focuses on the construction and evaluation of a hybrid CNN-LSTM-Attention model for land use forecasting and agricultural investment decision-making. The model is compared against traditional machine learning algorithms such as Random Forest, Support Vector Machine, and Gradient Boosting Machine to evaluate performance. In the experimental part, multi-dimensional data from agricultural zones of China were collected, including land use type, climate data, soil conditions, and crop yield (measured as annual crop production per hectare) from 2015 to 2020. The dataset of 10,000 samples spans the Eastern, Southern, and Western agricultural zones of China and is sourced from national agricultural surveys and publicly available environmental databases. The performance of the CNN-LSTM-Attention model was evaluated alongside the baseline models, with results showing that the CNN-LSTM-Attention model outperforms Random Forest, SVM, and GBM in land use change forecasting, achieving an accuracy of 96.8%. The study demonstrates the effectiveness of hybrid machine learning models for optimizing land use and making more accurate agricultural investment decisions. Additionally, the machine learning model predicted an average annual return of 12% for the best agricultural investment portfolio. In terms of agricultural investment decision-making, by combining land use forecasting and crop return data, the machine learning model successfully predicted the expected rate of return under different portfolios, with the best portfolio having an average annual return of 12%. This study shows that machine learning algorithms can effectively optimize land use structure and provide accurate predictions for agricultural investment decisions. The research results not only provide new ideas for the sustainable utilization of land resources, but also provide data support and decision-making basis for agricultural investors, and promote the development of agricultural modernization and intelligence.