Machine Learning and Deep Learning Architectures in Sustainable Agriculture: A Review of Crop Recommendation and Market Price Forecasting

Abstract ―​

Smallholder farmers in intensive agricultural regions face persistent ecological degradation and economic vulnerability caused by continuous monocropping of chemically intensive cash crops. In semi-arid regions such as Western Maharashtra, repeated cultivation of single crops like sugarcane and cotton has resulted in critical soil degradation, characterized by soil organic carbon depletion, secondary salinization, and microbial exhaustion. While restorative crop rotation with legumes, pulses, and root crops offers a natural biological remedy, smallholder adoption remains low. Farmers frequently avoid crop diversification due to high market price volatility in regional mandis, unpredictable post-harvest processing returns, and limited local access to certified organic amendments. Early diagnosis of soil fertility paired with dependable market forecasts can help farming communities understand individual land requirements and make informed, lower-risk agronomic choices. This review synthesizes four foundational studies by Bond and Lembke (2019), Pudumalar et al. (2017), Sendilkumar et al. (2022), and Hemamalini and Sundar (2022), focusing on soil degradation dynamics, machine learning crop classification, time-series price forecasting, and spatial logistics for bio-input distribution. The reviewed literature indicates that ensemble machine learning models, specifically Random Forest classifiers, reliably map multi-nutrient soil test indicators to restorative crops, while Long Short-Term Memory (LSTM) neural networks capture non-linear seasonal commodity price patterns better than classical econometric models. However, critical gaps remain: existing systems analyze soil chemistry and market prices in isolation, overlook post-harvest value-addition margins, and lack integration with localized, verifiable organic input supply chains. The review concludes that an integrated, farmer-centred, and data-driven approach can provide better economic and ecological opportunities for smallholder farmers.