<p>Hey devs! 👋</p>
<p>I just wrapped up a super interesting project for my ML Zoomcamp cohort and wanted to share the entire journey—wins, fails, and everything in between.</p>
<h2>What I Built</h2>
<p>An AI-powered crop yield prediction system that:</p>
<ul>
<li>Predicts harvest yields with 91% accuracy</li>
<li>Serves predictions via REST API</li>
<li>Runs on Google Cloud Run</li>
<li>Has a beautiful web UI</li>
</ul>
<h2>The Stack</h2>
<pre><code><span class="hljs-params">Backend:</span>  Python <span class="hljs-operator">+</span> Flask <span class="hljs-operator">+</span> Scikit-learn
<span class="hljs-params">DevOps:</span>   Docker <span class="hljs-operator">+</span> Google Cloud Run
<span class="hljs-params">Frontend:</span> Vanilla JS <span class="hljs-operator">+</span> HTML<span class="hljs-symbol">/CSS</span> (keeping it simple<span class="hljs-operator">!</span>)
<span class="hljs-params">ML:</span>       Gradient Boosting Regressor
</code></pre>
<h2>The Journey (Story Time! 📖)</h2>
<h3>Days 1-3: Data Exploration Hell 😅</h3>
<p>Started with messy agricultural data. Spent days just cleaning and understanding it. Pro tip: ALWAYS look at your data distributions first!</p>
<h3>Days 4-6: Model Selection Drama 🤖</h3>
<p>Trained 7 models. Linear Regression? Terrible. Decision Trees? Overfitting. Random Forest? Better but slow. Gradient Boosting? <em>Chef’s kiss</em> 👌</p>
<p>Here’s the comparison:</p>
<pre><code class="language-python">Model               | R² Score | MAE
--------------------|----------|--------
Gradient Boosting   | <span class="hljs-number">0.913</span>    | <span class="hljs-number">0.31</span>
Random Forest       | <span class="hljs-number">0.895</span>    | <span class="hljs-number">0.35</span>
Linear Regression   | <span class="hljs-number">0.623</span>    | <span class="hljs-number">0.89</span>
</code></pre>
<h3>Days 7-9: Docker Nightmares 🐳</h3>
<p>“Works on my machine” → Real problem.</p>
<p>Issue #1: Model files not loading in container
Solution: Load model at module level, not in <code>if __name__ == '__main__'</code></p>
<p>Issue #2: CORS blocking requests
Solution: <code>pip install flask-cors</code> saved my life</p>
<h3>Days 10-12: Cloud Deployment Victory! ☁️</h3>
<p>Google Cloud Run = Amazing for ML models</p>
<ul>
<li>Serverless (scales to zero!)</li>
<li>Easy Docker deployment</li>
<li>Built-in HTTPS</li>
<li>Pay per request</li>
</ul>
<h2>Code Snippets</h2>
<p>Here’s the prediction endpoint (simplified):</p>
<pre><code class="language-python"><span class="hljs-meta">@app.route(<span class="hljs-params"><span class="hljs-string">&#x27;/predict&#x27;</span>, methods=[<span class="hljs-string">&#x27;POST&#x27;</span>]</span>)</span>
<span class="hljs-keyword">def</span> <span class="hljs-title function_">predict</span>():
    data = request.get_json()
    
    <span class="hljs-comment"># Preprocessing</span>
    input_df = pd.DataFrame([data])
    input_encoded = pd.get_dummies(input_df)
    input_scaled = scaler.transform(input_encoded)
    
    <span class="hljs-comment"># Prediction</span>
    prediction = model.predict(input_scaled)[<span class="hljs-number">0</span>]
    
    <span class="hljs-keyword">return</span> jsonify({
        <span class="hljs-string">&#x27;predicted_yield&#x27;</span>: <span class="hljs-built_in">round</span>(prediction, <span class="hljs-number">2</span>),
        <span class="hljs-string">&#x27;unit&#x27;</span>: <span class="hljs-string">&#x27;tons_per_hectare&#x27;</span>
    })
</code></pre>
<h2>Biggest Learnings</h2>
<ol>
<li><strong>Data &gt; Models</strong>: Feature engineering mattered more than model selection</li>
<li><strong>Deployment is Hard</strong>: Spend time on DevOps early</li>
<li><strong>UI Matters</strong>: Built a simple HTML interface—users loved it</li>
<li><strong>Documentation</strong>: Write it as you code, not after!</li>
</ol>
<h2>Try It Yourself</h2>
<p>🔗 Live Demo: <a href="https://crop-yield-ui.web.app/" target="_blank" rel="noopener nofollow">Crop Yield Prediction API</a>
💻 GitHub: <a href="https://github.com/sojiadeyanju/mlzoomcamp_2025-midterm-project" target="_blank" rel="noopener nofollow">Github Repo</a></p>
<h2>What’s Next?</h2>
<p>Planning to add:</p>
<ul>
<li>[ ] Time-series forecasting</li>
<li>[ ] Weather API integration</li>
<li>[ ] Mobile app</li>
<li>[ ] Model retraining pipeline</li>
</ul>
<h2>Questions?</h2>
<p>Drop them in the comments! Happy to discuss anything about ML deployment, Docker, or agricultural AI! 👇</p>
<p>#MachineLearning #Python #DevOps #API</p>


Austin Deyan

A developer shares their 12-day journey building and deploying a crop yield prediction API using Python, Flask, and scikit-learn. The project achieved 91% accuracy using Gradient Boosting Regressor after testing multiple models. Key challenges included data cleaning, Docker containerization issues (model loading and CORS), and deployment to Google Cloud Run. The article includes code snippets, model performance comparisons, and practical lessons about feature engineering, deployment complexity, and documentation.

🌾 How I Built & Deployed a Crop Yield Prediction API in 12 Days