🤖 AIML Assignment — Uber Fare Prediction & Diabetes KNN Classification

Applied Machine Learning & AI coursework implementing regression and classification pipelines end-to-end.

---

📁 Repository Structure

├── Assignment1_Uber_Fare_Prediction.ipynb   # Regression: Uber fare prediction
├── Assignment2_KNN_Diabetes.ipynb           # Classification: Diabetes KNN
├── README.md
└── outputs/                                 # Generated plots (auto-created by notebooks)
    ├── boxplot_before.png
    ├── boxplot_after.png
    ├── correlation_heatmap.png
    ├── feature_importance.png
    ├── actual_vs_predicted.png
    ├── confusion_matrix.png
    ├── roc_curve.png
    └── ...

---

📌 Assignment 1 — Uber Fare Price Prediction

Problem Statement

Predict the fare amount for an Uber ride given pickup/drop-off coordinates, time, and passenger information.

Dataset

Uber Fares Dataset — 200,000 trips from New York City.

Column	Description
fare_amount	🎯 Target — Fare in USD
pickup_datetime	Timestamp of the ride
pickup_longitude/latitude	GPS coordinates of pickup
dropoff_longitude/latitude	GPS coordinates of drop-off
passenger_count	Number of passengers

Methodology

1. Pre-processing

• Parsed pickup_datetime → extracted hour, day_of_week, month, year
• Engineered binary features: is_weekend, is_rush_hour
• Computed Haversine distance (great-circle distance in km) between pickup and drop-off
• Applied business logic filters (valid NYC coordinates, fare > $0, etc.)

2. Outlier Detection

Used the IQR (Interquartile Range) method:

Lower bound = Q1 − 1.5 × IQR
Upper bound = Q3 + 1.5 × IQR

Removed outliers from fare_amount and distance_km.

3. Correlation Analysis

• Computed Pearson correlation matrix
• Visualised as a heatmap
• distance_km shows the strongest positive correlation with fare

4. Models

Model	Description
Linear Regression	Baseline — assumes linear relationship between features and fare
Random Forest Regressor	Ensemble of 100 trees — captures non-linear patterns

Results

Metric	Linear Regression	Random Forest
MAE	Higher	Lower ✅
RMSE	Higher	Lower ✅
R²	Lower	Higher ✅

Winner: Random Forest — captures the non-linear relationship between distance, time, and fare.

---

📌 Assignment 2 — K-Nearest Neighbors on Diabetes Dataset

Problem Statement

Classify whether a patient has diabetes based on diagnostic measurements. Evaluate the model using a full set of classification metrics.

Dataset

Diabetes Dataset (Pima Indians) — 768 female patients.

Feature	Description
Pregnancies	Number of times pregnant
Glucose	Plasma glucose concentration
BloodPressure	Diastolic blood pressure (mm Hg)
SkinThickness	Triceps skin fold thickness (mm)
Insulin	2-Hour serum insulin (mu U/ml)
BMI	Body mass index
DiabetesPedigreeFunction	Diabetes pedigree function
Age	Age in years
Outcome	🎯 Target — 1 = Diabetes, 0 = No Diabetes

Methodology

1. Pre-processing

• Replaced biologically impossible zero values (Glucose, BMI, etc.) with class-conditional medians
• Applied StandardScaler — essential for KNN since it is distance-based

2. Finding Optimal K

Used the Elbow Method — plotted error rate vs K (1–30) to find the K that minimises error.

3. KNN Algorithm

For each test point:
  1. Compute Euclidean distance to all training points
  2. Select K nearest neighbours
  3. Majority vote → predicted class

4. Evaluation Metrics

Metric	Formula	Meaning
Accuracy	(TP+TN) / Total	Overall correctness
Error Rate	1 − Accuracy	Overall mistakes
Precision	TP / (TP+FP)	Of predicted +ve, how many are correct?
Recall	TP / (TP+FN)	Of actual +ve, how many did we catch?
F1-Score	2·P·R / (P+R)	Harmonic mean of Precision & Recall
AUC-ROC	Area under curve	Overall discrimination ability

Results Summary

Metric	Score
Accuracy	~77–80%
Precision	~70–75%
Recall	~60–68%
F1-Score	~65–72%
AUC-ROC	~80–84%

---

🚀 How to Run

Option A — Google Colab (Recommended)

1. Open Google Colab
2. Click File → Upload Notebook → upload .ipynb file
3. Upload your dataset when prompted (or use the built-in synthetic/sklearn loader)
4. Click Runtime → Run All

Option B — Local Jupyter

# 1. Clone the repository
git clone https://github.com/YOUR_USERNAME/aiml-assignment.git
cd aiml-assignment

# 2. Install dependencies
pip install pandas numpy matplotlib seaborn scikit-learn jupyter

# 3. Launch Jupyter
jupyter notebook

Datasets

Download from Kaggle and place in the same directory as the notebooks:

• Assignment 1: uber.csv
• Assignment 2: diabetes.csv

Note: Both notebooks include fallback data loaders so they run without the Kaggle files (using sklearn's built-in Pima dataset for A2 and synthetic data for A1).

---

🛠️ Tech Stack

Tool	Purpose
Python 3.10+	Core language
Pandas	Data manipulation
NumPy	Numerical operations
Matplotlib / Seaborn	Visualisation
scikit-learn	ML models & evaluation
Google Colab	Cloud execution

---

📊 Key Visualisations Generated

Assignment 1:

• Boxplots before/after outlier removal
• Fare amount distribution
• Correlation heatmap
• Fare vs Distance scatter plot
• Average fare by hour of day
• Random Forest feature importance
• Actual vs Predicted plots
• Residual plots

Assignment 2:

• Feature distributions by outcome class
• Correlation heatmap
• Elbow method (Error Rate vs K)
• Confusion matrix (counts + normalised)
• ROC curve with AUC
• Metrics dashboard
• KNN decision boundary (Glucose vs BMI)

---

📤 Push to GitHub

# First time setup
git init
git add .
git commit -m "Add AIML assignments: Uber fare prediction + Diabetes KNN"
git branch -M main
git remote add origin https://github.com/YOUR_USERNAME/aiml-assignment.git
git push -u origin main

# Subsequent updates
git add .
git commit -m "Update notebooks"
git push

---

👤 Author

Devendra Agrawal
Department of Computer Engineering / AI & ML
MIT World Peace University | 2025-27

---

📄 License

MIT — free to use and modify for educational purposes.