Agricultural land suitability classification and crop suggestion using machine learning and spatial multicriteria decision analysis in semi-arid ecosystem

Agricultural land evaluation for crop cultivation plays a vital role in developing a sustainable and productive agricultural system. The traditional way of manual land evaluation is time-consuming, costlier, requires expert knowledge, and prone to human errors. The application of machine learning is making its impact on almost every field. To overcome the challenges of manual land evaluation automated approach of machine learning can play an influential role in agriculture. Therefore, the present study aimed to investigate and propose an automated solution for agricultural land suitability evaluation for two major crops, i.e., wheat and mustered, based on a hybrid approach of multicriteria decision analysis (MCDA) and machine learning (ML). The proposed framework is based on the land suitability assessment framework proposed by the Food and Agriculture Organization (FAO). For evaluating suitability levels of land units, three major influential categories of geospatial parameters are considered, i.e., soil, climatic, and topographic. The MCDA technique of analytical hierarchy process (AHP) has been used to analyze crop suitability utilizing these geospatial parameters. The ML models are constructed to map the quantified crop suitability results and suggest the most suitable crop to be cultivated. The obtained results indicate that the Balanced Bagging classifier achieved the highest balanced accuracies of 97.69% and 97.57% for mustered and wheat crop suitability predictions, respectively, and a balanced accuracy of 97.36% for crop classification. Since the AHP method is a systematic approach to decision analysis, while ML is a data-driven approach, the proposed unified approach offers more comprehensive predictions. The potential of the proposed framework is evaluated for the Abhanpur district of Chhattisgarh state, India. The proposed ML-based approach overcomes the limitations of the conventional methods and provides precise decisions; therefore, could turn the attention to ML in future cropland suitability evaluation studies.