Back

SoilMate

Helping farmers swap guesswork for data.

PythonMachine LearningStreamlitPandasData CleaningDashboards
Year
2025
Context
MSc group project · University College Cork
My role
6-person group. I extracted and cleaned the data, built the Streamlit dashboard, and designed the system architecture.
Links
Code available on GitHub

The Problem

In India, most small farmers choose crops by tradition and word of mouth, not data. Information about soil, rainfall and market prices exists, but it's spread across different government websites and hard to use. The result is wrong crops, wasted water, and unpredictable income.

The Approach

  1. 1

    Collect the data

    Soil, weather, market prices and crop information from Indian government websites, covering 2018–2024 across the state of Karnataka.

  2. 2

    Clean it up

    The same district was spelled differently in different files, and some readings were missing. We fixed the names and filled the gaps so everything could be joined into one reliable dataset.

  3. 3

    Turn numbers into signals farmers care about

    Like a soil quality score and a water stress score, instead of raw chemistry readings.

  4. 4

    Test models, keep the best

    We tried four different prediction methods and kept the one that picked the right crop most often: 92.8% of the time.

  5. 5

    Build the dashboard

    A simple five-tab app: a farmer picks their district, season and water source, and instantly sees what to grow, what it costs, and what it could earn.

The Evidence

Pick your district and season to get a crop recommendation with the reasons why.
Pick your district and season to get a crop recommendation with the reasons why.
How data flows: sources → cleaning → predictions → dashboard.
How data flows: sources → cleaning → predictions → dashboard.
We tested four methods and kept the most accurate one.
We tested four methods and kept the most accurate one.
What it costs, what it earns, before planting a single seed.
What it costs, what it earns, before planting a single seed.

The Outcome

0.0%

picked the right crop in testing

0%+

accuracy of cost and income forecasts

0

farm situations covered, with instant answers

Next project

Predicting Which Dublin Properties Will Sell