AI Assistant Usage in Student Life — Classification Project

This project analyzes synthetic data of AI assistant usage among students, performs exploratory data analysis (EDA), feature engineering, preprocessing, and builds a classification model to predict whether a student will use the AI assistant again in the future.

---

📂 Dataset

The dataset is publicly available on Kaggle:

AI Assistant Usage in Student Life (Synthetic)

It contains 10,000 records with the following columns:

• SessionID — Unique identifier for each AI assistant usage session
• StudentLevel — Education level (e.g., Undergraduate, Graduate, High School)
• Discipline — Field of study (e.g., Computer Science, Psychology)
• SessionDate — Date of the AI assistant session
• SessionLengthMin — Duration of session in minutes
• TotalPrompts — Number of prompts sent during the session
• TaskType — Type of task performed (e.g., Studying, Coding)
• AI_AssistanceLevel — Level of AI assistance (1 to 5)
• FinalOutcome — Outcome of the session (e.g., Assignment Completed)
• UsedAgain — Target variable (1 = Yes, 0 = No)
• SatisfactionRating — Satisfaction rating given by the student

---

📊 Project Workflow

1. EDA (Exploratory Data Analysis)

• Checked dataset shape, data types, and missing values
• Summary statistics (describe())
• Class distribution visualization for the target variable
• Distribution plots for numeric variables
• Count plots for categorical variables

2. Feature Engineering

Additional features were created to improve model performance:

• Date-based features extracted from SessionDate:
  • year
  • month
  • dayofweek
  • day
  • weekofyear
• Interaction-based feature:
  • prompts_per_min = TotalPrompts / SessionLengthMin

3. Preprocessing

• Encoding categorical variables with LabelEncoder (for simplicity)
• Scaling numeric features using StandardScaler
• SMOTE applied to handle class imbalance
• Train-test split (80/20 ratio)

4. Modeling

Models tested:

• Logistic Regression
• Random Forest Classifier
• XGBoost Classifier

Final chosen model: RandomForestClassifier

• Achieved accuracy: ~75%

5. Evaluation

• Confusion Matrix
• Classification Report (Precision, Recall, F1-score)
• Accuracy Score
  imbalanced-learn xgboost