A Random Forest regression model predicting lettuce growth duration from environmental sensor data — built, tuned, and deployed as an interactive prediction tool.
Develop a complete data science workflow on a real-world dataset — exploratory data analysis, a data-driven narrative, and machine-learning-based predictive modelling — built iteratively in a shared Google Colab notebook over a 6-week project timeline.
Validated data types — converting dates correctly — and identified and corrected invalid sensor readings, including pH readings above 14 and negative humidity values. Used box plots against growth duration to detect outliers before any modelling began.
Temperature turned out stable and low-impact, which made sense given the greenhouse-controlled environment. TDS and pH — particularly within mid-range levels — were the most influential variables. No single variable showed a strong linear relationship with growth duration on its own, which is what motivated a multivariate machine learning approach rather than simple regression.
The core question: can we estimate plant growth duration from environmental data? Three models were tested —
Produced an actual-vs-predicted scatter plot, a residual distribution plot confirming small, near-normal errors, and a feature importance chart — with pH and TDS ranking as the top predictors, consistent with the earlier exploratory findings.
Built an interactive prediction tool using ipywidgets — a user inputs temperature, humidity, and pH and receives an estimated growth duration, turning the analysis into something an actual hydroponic grower could use, not just a notebook result.
Notebook screenshots not yet added — send the final Colab export and this page gets the real charts.