Deepak Srivastava, Vibhor Sharma, Sameer Chavan, Suman Pant and Ashutosh Kumar Choudhary
Abstract
Irregular monsoon initiation, fragmented soil quality over small geographic regions, and limited access to timely and targeted agronomic guidance are recurrent difficulties for smallholder farmers in the Himalayan foothills of Uttarakhand. Most current crop recommendation algorithms only work on a regional scale, failing to take district-level meteorological factors or planting limitations into consideration. This research presents Fasal Mitra, an intelligent agricultural advisory system that integrates a district-level knowledge base, a natural language advising module, and ensemble machine learning for crop prediction. Using a benchmark soil-climate dataset consisting of 2,200 samples and seven agronomic variables, we created and evaluated six classification algorithms. Classification accuracy on the held-out test set was 100% for Random Forest, Support Vector Machine with standard scaling, and Logistic Regression with standard scaling. Sowing schedules, yield benchmarks, pest and disease profiles, and time-indexed contingency rules for unfavourable weather events like drought and flood are all part of the systemâs district knowledge baseâs organized agronomical data for five districts in Uttarakhand. Using a conversational interface that is based on keyword-grounded knowledge retrieval, farmers may ask for district-specific help using a natural language advising module. For users in more remote areas, the system is available as an online app. Future work will integrate real-time weather forecasts, soil sensor measurements, and remote sensing data to further enhance climate-resilient decision support, optimize resource use, and strengthen environmental sustainability across broader agricultural regions.