Start with your problem type, not the algorithm. What are you trying to predict?

Problem Type = Algorithm Type

Predicting categories? (spam/not spam, cat/dog)

→ Classification algorithms:
• Logistic Regression (simple, fast)
• Random Forest (good default)
• Neural Networks (complex data)

Predicting numbers? (price, sales, temperature)

→ Regression algorithms:
• Linear Regression (simple relationships)
• Random Forest Regressor (non-linear)
• Gradient Boosting (competitions)

Finding groups? (customer segments)

→ Clustering algorithms:
• K-Means (simple, fast)
• DBSCAN (odd shapes)

Decision Shortcuts

Small data (< 10k rows): Start with Logistic/Linear Regression

Medium data + accuracy matters: Random Forest or Gradient Boosting

Images/text/audio: Neural Networks (deep learning)

Need explainability: Decision Trees, Linear models

The Practical Approach

1. Start simple (Logistic/Linear Regression)
2. Benchmark with Random Forest
3. Try Gradient Boosting if you need more accuracy
4. Only use neural networks if data is complex (images, text)