Introduction: The Power of Data Science

Data science is a cornerstone of modern decision-making. It helps organizations uncover hidden patterns, forecast outcomes, and make data-informed decisions using a mix of statistics, programming, and domain knowledge.

But what if you could do all this—without writing code?

That’s where TernoAI comes in. It's a powerful AutoML tool that takes natural language prompts and turns them into complete data science workflows, from EDA to model evaluation.

Project Setup: Uploading Kaggle's Medical Cost Dataset to TernoAI

To begin this analysis, I manually downloaded the Medical Cost Personal Dataset from Kaggle, which contains 1,338 records of individual health profiles alongside their annual medical insurance charges.

After downloading the insurance.csv file, I uploaded it directly to TernoAI.

About the Dataset

The dataset includes the following features:

Column	Description
age	Age of the individual
sex	Gender (male/female)
BMI	Body Mass Index
children	Number of dependents covered
smoker	Smoking status (yes/no)
region	Region of residence (southeast, northwest, etc.)
charges	Annual medical insurance cost (target variable)

With just one upload, the dataset was ready for immediate exploration—no extra setup, no dependency headaches.

The Challenge: Predicting Medical Insurance Costs

With the data successfully uploaded into TernoAI, the core objective was to build a machine-learning model that could predict annual medical insurance costs based on individual attributes like age, BMI, smoking status, and region.

This task isn’t just about prediction—it also offers opportunities to:

• Identify cost drivers

• Enable personalized premium strategies.

• Guide public health interventions based on behavioral patterns (e.g., smoking)

Using Terno AI: From Command to Workflow

Instead of writing code manually, I used Terno AI, a smart AutoML assistant that responds to plain English commands. My first prompt was:

“Perform exploratory data analysis on the Medical Cost Personal Dataset, including summary statistics, missing value analysis, duplicate value analysis, and visualizations for key features.”

From this single prompt, TernoAI instantly delivered:

• Cleaned and analyzed the dataset

• Handled missing/duplicate checks

• Generated insightful visualizations

• Prepared the data for modeling

Exploratory Data Analysis (EDA)

Summary Statistics

TernoAI computed key stats for numeric features:

Feature	Mean	Std Dev	Min	Max
Age	39.2	14.0	18	64
BMI	30.7	6.1	15.96	53.13
Charges	13,270	12,100	1,122	63,770

Insight: Medical charges exhibit high variance and are skewed to the right, indicating that a few individuals incur significantly higher costs.

Missing & Duplicate Values

TernoAI reported:

• Missing Values: None

• Duplicate Rows: 1 duplicate found and handled
  The dataset was clean and ready for analysis.

Visual Insights

• Distributions and Categorical Counts
  This figure provides an overview of the data distribution across both numerical and categorical features.

• Histograms show the shape and skewness of continuous variables like age, BMI, and charges.

• Count plots reveal the balance across sex, smoker, and region.
  Notably, charges are highly skewed, and the dataset has more non-smokers than smokers.

• Relationships Between Features and Medical Charges
  This figure explores how smoker status and BMI relate to insurance charges.

• The boxplot shows that smokers have substantially higher charges, confirming that smoking is a key cost driver.

• The scatterplot of BMI vs charges, colored by smoker, reveals that higher BMI correlates with higher costs, especially among smokers.

Model Building and Evaluation

Before diving into modeling, I wanted to evaluate which machine learning algorithms could best distinguish between high-cost and low-cost individuals using the available features.

My Prompt:

“Train the best models that you think work best for this data, handle the missing values, encode the categorical variables and also evaluate the model performance using the accuracy, precision, recall and F1-score, make a table for all scores to compare. Then, at last, compare all the models and pick the best according to performance, and show the inference.”

Problem Framing

To simplify prediction, TernoAI transformed the problem into a binary classification task:

• 1 = High-cost individual

• 0 = Low-cost individual
  This made it easier to train classification models that predict cost categories, a valuable feature for insurers.

TernoAI built complete training pipelines for the following models:

• Logistic Regression

• Decision Tree

• Random Forest

• Gradient Boosting

All preprocessing steps were automated:

• One-hot encoding

• Scaling

• Train-test splitting (80/20)

Model Evaluation Results

Model	Accuracy	Precision	Recall	F1-Score
Logistic Regression	0.8993	0.8849	0.9179	0.9011
Decision Tree	0.8806	0.8806	0.8806	0.8806
Random Forest	0.9366	0.9680	0.9030	0.9344
Gradient Boosting	0.9291	0.9528	0.9030	0.9272
Best Model: Random Forest

• Highest accuracy and F1-score**
  **

• Excellent tradeoff between precision and recall

Visual Comparison of Model Performance

TernoAI generated a grouped bar chart to compare performance metrics visually.

According to the chart, Random Forest outperforms across all four metrics.

Feature Correlation Analysis

To gain deeper insight into what drives medical costs, I asked:

My Prompt:

“So what is your conclusion on what features have the best correlation with medical insurance and which have the worst?”

TernoAI computed Pearson correlations between features and charges:

Top Correlated Features:

Feature	Correlation
Smoker (yes)	+0.787
Age	+0.299
BMI	+0.198
Children	+0.067
Region/Sex	~0.00
Interpretation:

• Smoking is the strongest single-cost predictor

• Age and BMI are moderate cost drivers.

• Sex and region have minimal impact.

Conclusion: What TernoAI Helped Me Achieve

With minimal input and zero manual coding, TernoAI helped me:

• Clean and explore the dataset

• Visualize important relationships

• Convert a regression problem into a classification problem.

• Train and evaluate four machine learning models.

• Identify top predictors of healthcare costs.

• Choose the best model (Random Forest) based on strong metrics.

Key Insights:

• Smoking is the most influential driver of medical costs.

• Random Forest was the best-performing model.

• TernoAI automated the entire analysis pipeline—cleanly and correctly.

Why You Should Try TernoAI

If you're:

• A beginner who wants to analyze data without coding

• A professional looking to automate repetitive tasks

• A data scientist who wants a second brain
  Then TernoAI is your AI co-pilot. With just a natural language command, I got full-scale machine learning analysis, reproducible code, visuals, and performance metrics—faster than ever before.

You just have to:

• Upload your dataset

• Type what you want

• Get results, models, code, and plots instantly
  No notebooks. No config files. Just results.

Chat source –https://nishtha.app.terno.ai/chat/share/89577b09-88c6-49cf-8979-b29486508cd1