Today I spent time reviewing introductory machine learning concepts, working through practice questions, and trying to understand the reasoning behind the answers instead of simply memorizing formulas.
One thing I realized is that many machine learning concepts become much easier when I think about what the numbers actually represent.
Classification vs. Regression
A simple way I remember the difference is:
- Classification predicts a category or label.
- Examples: High Risk vs. Low Risk, Hospital Readmission (Yes/No).
- Regression predicts a continuous value.
- Examples: Annual healthcare cost, Hospital length of stay.
For me:
- 🌳 Decision Trees = Classification.
- 📈 Linear Regression = Regression.
Confusion Matrix, Accuracy, and Precision
I found it easier to think about a confusion matrix visually rather than memorizing definitions.
- Accuracy asks: How many predictions were correct overall?
- Precision asks: Of all the cases predicted as positive, how many were actually positive?
- Recall (Sensitivity) asks: Of all the truly positive cases, how many did the model correctly identify?
A simple memory trick I wrote down:
- Accuracy = Correct predictions ÷ Total predictions.
- Precision = True Positives ÷ Predicted Positives.
- Recall = True Positives ÷ Actual Positives.
Cross Validation
I also reviewed k-fold cross validation.
For example, if there are 1,000 records and we use 5-fold cross validation:
- Each fold contains 200 records.
- One fold is used for testing.
- The remaining four folds (800 records) are used for training.
Thinking of it as “1 part testing and 4 parts training” made it much easier to remember.
Linear Regression
Linear regression predicts a number using a best-fit line.
The equation is usually written as:
y = ax + b
where:
- a = slope,
- b = intercept.
One interesting thing I learned is that the regression line does not have to pass through every data point exactly. It is simply the line that best fits all the data overall.
Entropy: A Simpler Way to Think About It
At first, entropy looked intimidating because of the formula. What helped me was ignoring the formula for a while and focusing on the idea.
Entropy measures how mixed or uncertain the data is.
I started using a simple mental picture:
| Data Distribution | What It Means |
|---|---|
| 100% / 0% | Completely pure |
| 99% / 1% | Almost pure |
| 50% / 50% | Completely mixed |
The closer the data is to 50/50, the higher the entropy. The closer it is to 100/0, the lower the entropy.
I do not try to memorize the full entropy equation. I just remember that:
- Pure data → Low entropy.
- Mixed data → High entropy.
Information Gain: The Concept That Finally Clicked
Information Gain was the most challenging topic for me today.
What finally helped was asking one simple question:
Did the split actually make the data more organized?
Instead of jumping directly into calculations, I looked at what happened after the split.
For example, if only one branch becomes perfectly organized while the other branches are still mixed, then the Information Gain should probably be moderate rather than very high.
I realized that my “guesses” were not really random. I was already thinking about how much uncertainty had been reduced. The formula simply turns that intuition into a number.
Connecting Machine Learning to Healthcare
As someone transitioning from healthcare coding and risk adjustment into healthcare data analytics, I find it much easier to understand machine learning by relating it to healthcare examples.
Classification examples:
- High-risk vs. low-risk patients.
- Predicting 30-day readmission.
- Identifying members who may benefit from care management.
Regression examples:
- Predicting annual healthcare costs.
- Estimating hospital length of stay.
- Forecasting future healthcare utilization.
Thinking about these real-world applications makes the concepts much easier to remember.
My Biggest Takeaway
Today I realized that I do not need to memorize every formula perfectly. What matters most is understanding the idea behind the formula.
- Entropy measures uncertainty.
- Information Gain measures how much uncertainty decreases after a split.
- Decision trees look for the split that creates the clearest groups.
- Regression predicts numbers.
- Classification predicts categories.
Once I understood what each concept was trying to measure, the formulas became much less intimidating. Sometimes building intuition first is the best way to learn.