🤖 DevelopersHub Corporation — AI/ML Engineering Internship

A portfolio of completed machine learning projects spanning data exploration, predictive modeling, and binary classification.

---

📋 Project Overview

#	Project Title	Dataset	Model(s)	Status
01	Dataset Exploration & Visualization — Iris	Iris (UCI)	EDA only	✅ Complete
02	Short-Term Stock Price Prediction — AAPL	Yahoo Finance via yfinance	Linear Regression	✅ Complete
03	Heart Disease Risk Prediction — UCI Cleveland	UCI Cleveland Heart Disease	Logistic Regression · Decision Tree	✅ Complete
06	House Price Prediction — Regression & Feature Analysis	Synthetic / Kaggle-structured	Linear Regression · Gradient Boosting	✅ Complete

---

📁 Project Structure

developershub-aiml-internship/
├── Task_1.ipynb               # Iris Dataset — EDA & Visualization
├── Task_2.ipynb               # AAPL Stock Price Prediction
├── Task_3.ipynb               # Heart Disease Risk Classification
├── Task_6.ipynb               # House Price Regression
├── README.md                  # This file
└── README.html                # Styled portfolio version

---

🔬 Project 01 — Dataset Exploration & Visualization

Objective: Load, inspect, and visualize the Iris dataset to understand data distributions and feature relationships.

Dataset: Iris Dataset — UCI Repository (loaded via seaborn)

What was done:

• Loaded dataset using pandas — printed shape, columns, .head(), .info(), and .describe()
• Scatter plot of sepal length vs sepal width colored by species, plus a full pairplot across all features
• Histograms for all 4 features with KDE overlays grouped by species
• Box plots per feature to identify outliers across all three species

Key Findings:

• Petal length and petal width are the most discriminating features between species
• Iris Setosa is clearly linearly separable from the other two species
• Versicolor and Virginica overlap in sepal dimensions but are separable via petal measurements

File: Task_1.ipynb

---

📈 Project 02 — Short-Term Stock Price Prediction

Objective: Use historical Apple (AAPL) stock data to predict the next day's closing price.

Dataset: Apple Inc. (AAPL) — Yahoo Finance via yfinance (January 2020 – January 2024)

Model: Linear Regression

Metric	Value
Features used	Open, High, Low, Volume
Target	Next-day Close price
Split strategy	shuffle=False — time series order preserved
Evaluation	MSE · RMSE

Key Findings:

• Linear Regression captures the general price trend well for next-day prediction
• Using shuffle=False is critical for time series data — prevents leakage from the future
• Today's Open/High/Low values are strong proxies for tomorrow's Close price

File: Task_2.ipynb

---

🫀 Project 03 — Heart Disease Risk Prediction

Objective: Predict whether a patient is at risk of heart disease using 13 clinical health features.

Dataset: UCI Heart Disease Dataset — Cleveland subset (242 clean records, 13 features, binary target)

Model Results:

Model	Accuracy	ROC-AUC
Logistic Regression	93.9%	0.982
Decision Tree (max_depth=5)	91.8%	0.942

What was done:

• Loaded UCI Cleveland dataset and handled ? missing values in ca and thal columns
• Binarised target: 0 = no disease, 1 = disease present
• Full EDA — class distribution, age histograms, max heart rate scatter, chest pain type analysis, cholesterol box plots
• Trained both models; evaluated with accuracy, confusion matrix, ROC curve, and AUC score
• Feature importance (Decision Tree) and coefficient magnitude (Logistic Regression) charts

Key Findings:

• ca (number of major vessels blocked) and cp (chest pain type) are the top two predictors
• oldpeak (ST depression) and thal (thalassemia type) are the next most important features
• Logistic Regression significantly outperforms Decision Tree on AUC — 0.982 vs 0.942
• Model achieves strong clinical screening performance, suitable as a patient triage support tool

File: Task_3.ipynb

---

🏠 Project 06 — House Price Prediction

Objective: Predict house prices using property features — square footage, bedrooms, location, age, and amenities.

Dataset: Synthetic dataset (1,500 records, 9 features) structured to mirror the Kaggle House Price dataset

Model Results:

Model	MAE	RMSE
Linear Regression	$33,665	$41,341
Gradient Boosting (200 estimators)	$24,374	$30,626

What was done:

• Generated a realistic dataset with location-based price premiums and natural Gaussian noise
• Preprocessing — Label Encoding for Location, StandardScaler applied for Linear Regression
• Full EDA — price distribution, price vs SqFt scatter by location, average price by location, correlation heatmap, box plots by bedrooms and garage spaces
• Trained both models; plotted actual vs predicted scatter with a perfect-fit reference line
• Feature importance bar chart from Gradient Boosting regressor

Key Findings:

• SqFt (square footage) is by far the strongest single predictor of house price
• Location is the second most important feature — Downtown/Uptown carry a ~$80,000 premium
• Gradient Boosting outperforms Linear Regression by 27% on MAE, capturing non-linear interactions
• Age has a consistent negative effect; Pool and Garage add measurable value premiums

File: Task_6.ipynb

---

⚙️ Setup & Installation

# Core dependencies — all projects
pip install pandas numpy matplotlib seaborn scikit-learn

# Project 02 — Stock data fetching
pip install yfinance

---

🧠 Skills Demonstrated

Category	Skills
Data Engineering	Data loading, cleaning, preprocessing, missing value handling
Visualization	Matplotlib, Seaborn — scatter plots, histograms, box plots, heatmaps
Regression	Linear Regression, Gradient Boosting Regressor
Classification	Logistic Regression, Decision Tree Classifier
Model Evaluation	MAE, RMSE, Accuracy, ROC-AUC, Confusion Matrix, Feature Importance
Time Series	Sequential train/test splits, preventing data leakage

---

🔗 Related Repository

Task 04 — General Health Query Chatbot (Prompt Engineering & LLM) is maintained in a separate repository. 👉 developershub-health-chatbot

---

👤 Author

Arman Adil Mangat AI/ML Engineering Intern — DevelopersHub Corporation