Speakers
Description
Floods, human-elephant conflict, crop damage, health problems, and extreme weather in Assam and this region do not happen alone — they are connected. Floodwater pushes elephants into farmland right when crops are ready to harvest, and dirty floodwater later affects people's health, making the next flood harder to deal with. We call this a coupled human-natural system. This work presents AssamWatch, which studies district-level risk across all 35 districts of Assam by combining three types of information: what people report during a hazard, real-time weather data, and a health risk score from local communities. To read community reports written in Assamese, English, or Bengali, we built a simple word-based method needing no labeled training data. It classified posts with a macro-F1 score of 0.919 (92.6% accuracy) on 81 test posts, clearly better than TF-IDF with SVM (0.583) or Logistic Regression (0.269). We also tested using only weather data: two districts with almost identical weather looked nearly the same (only 2.0 points apart), but adding the other two layers back made the gap much clearer (11.2 points apart), showing weather data alone is not enough. The health score was also meaningfully linked to student performance in colleges (p = 0.007). Every result comes with a simple explanation, and the system already runs for free online. This study is a real example of studying a connected, nonlinear system using data and AI methods.