Soil Nutrient Level Prediction using Linear Regression in ML

Soil scientists and agronomists often need a quick estimate of the primary macronutrients—nitrogen (N), phosphorus (P), and potassium (K)—present in a soil sample so they can prescribe fertiliser rates and plan crop rotations.

In this mini‑project, we build a linear‑regression baseline that predicts each nutrient’s concentration (parts‑per‑million) from inexpensive, easy‑to‑obtain field measurements: soil pH, electrical conductivity, organic‑matter percentage, moisture, and region. A transparent line highlights the factors that most strongly influence N, P, and K, and provides a benchmark before switching to more complex regressors.

Libraries Required

• pandas # data wrangling
• numpy # numerical helpers
• matplotlib.pyplot # optional quick plots
• scikit‑learn # preprocessing, model, metrics
• joblib # save the trained pipeline

Dataset Link

Soil measures

Step-by-Step Code Implementation

Why linear regression first? In many soils, macronutrient levels respond nearly linearly to pH, organic matter, and salinity within normal ranges. A straight‑line fit quantifies these elasticities and is trivial to explain to extension agents.

1. Import Libraries

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

from sklearn.model_selection import train_test_split
from sklearn.preprocessing import OneHotEncoder, StandardScaler
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from sklearn.linear_model import LinearRegression
from sklearn.metrics import r2_score, mean_absolute_error
import joblib

2. Load & peek at the data

Download soil_measures.csv from the given link above:

df = pd.read_csv("soil_measures.csv")
print(df.head())

Typical columns

column	description
N / P / K	macronutrient ppm (targets)
pH	soil pH (0–14)
EC	electrical conductivity (dS m⁻¹)
OrganicMatter	% by weight
Moisture	% at sampling time
Region	categorical (e.g., North, East …)

3. Basic cleaning

core = ['N','P','K','pH','EC','OrganicMatter','Moisture','Region']
df   = df.dropna(subset=core).copy()

4.  Choose predictors & label(s)

We will train one model per nutrient to ensure crystal-clear interpretation.

num_cols = ['pH', 'EC', 'OrganicMatter', 'Moisture']
cat_cols = ['Region']

5.  Reusable preprocessing pipeline

Standard scaling and one-hot encoding keep coefficients comparable (ppm per 1 σ) and treat Region as categorical, avoiding fake numeric ordering.

preproc = ColumnTransformer([
        ('cat', OneHotEncoder(handle_unknown='ignore'), cat_cols),
        ('num', StandardScaler(),                      num_cols)
])

linreg = LinearRegression()
pipe   = Pipeline([('prep', preproc),
                   ('model', linreg)])

6. Train, evaluate, and save for each nutrient

• Separate models for N, P, and K avoid multivariate target interactions that could complicate interpretation.
• Evaluation metrics – R² indicates the percentage of nutrient variance captured; MAE in ppm provides a tangible error band. Agronomists can decide if ±3 ppm is “good enough” for fertiliser planning.
• Joblib persistence lets a mobile soil‑testing app load the .pkl files and output nutrient estimates on‑site without retraining.

models = {}
for nutrient in ['N', 'P', 'K']:
    X = df[num_cols + cat_cols]
    y = df[nutrient]

    X_train, X_test, y_train, y_test = train_test_split(
            X, y, test_size=0.2, random_state=42)

    pipe.fit(X_train, y_train)
    pred = pipe.predict(X_test)

    print(f"\n=== {nutrient} Model ===")
    print(f"R²  : {r2_score(y_test, pred):.3f}")
    print(f"MAE : {mean_absolute_error(y_test, pred):.1f} ppm")

    models[nutrient] = pipe
    joblib.dump(pipe, f"soil_{nutrient.lower()}_linreg.pkl")

7. Interpreting coefficients (example for Nitrogen)

Coefficient tables reveal actionable levers, such as a substantial positive weight on OrganicMatter for N, suggesting that adding compost could boost nitrogen reserves.

ohe_feats = models['N'].named_steps['prep']\
                       .named_transformers_['cat']\
                       .get_feature_names_out(cat_cols)
all_feats = list(ohe_feats) + num_cols

coef_series = pd.Series(models['N'].named_steps['model'].coef_,
                        index=all_feats).sort_values()

print("\nFactors decreasing N content:")
print(coef_series.head(5))

print("\nFactors increasing N content:")
print(coef_series.tail(5))

Because inputs are z‑scored, each coefficient reads as ppm change in N for a one‑standard‑deviation change in that feature.

Summary

In roughly 100 lines of Python, we transformed routine soil‑lab readings into an explainable nutrient‑content forecaster:

• Instant ppm estimates help farmers fine‑tune fertiliser rates before planting.
• Transparent coefficients precisely quantify how pH, salinity, organic matter, and moisture influence N-P-K levels.

Keep this linear baseline as a benchmark—when you switch to tree ensembles or spectroscopy‑based models, you’ll know precisely how much additional predictive value the sophistication brings.