This project aims to predict the performance of students based on various demographic and educational factors. The goal is to use regression analysis to predict student scores in mathematics.

The dataset used for this project includes the following variables:

• gender: Gender of the student (male/female)
• race_ethnicity: Ethnic group of the student
• parental_level_of_education: Highest education level of the student's parents
• lunch: Type of lunch (standard/free/reduced)
• test_preparation_course: Completion status of the test preparation course (completed/none)
• math_score: Student's score in mathematics
• reading_score: Student's score in reading
• writing_score: Student's score in writing

• Overall performance score: A combined metric based on math, reading, and writing scores

The dataset can be accessed from Kaggle.

It is observed that the categorical variables 'gender', 'race_ethnicity', 'parental_level_of_education', 'lunch', and 'test_preparation_course' play significant roles in predicting student performance.

• In the Data Ingestion phase, the data is first read as a CSV file.
• Then the data is split into training and testing sets and saved as CSV files.

• In this phase, a ColumnTransformer pipeline is created.
• For numeric variables, first, SimpleImputer is applied with the median strategy, then standard scaling is performed on the numeric data.
• For categorical variables, SimpleImputer is applied with the most frequent strategy, then ordinal encoding is performed, and after this, the data is scaled with StandardScaler.
• This preprocessor is saved as a pickle file.

• In this phase, various regression models are tested. The performance of these models is evaluated based on their R2 scores.
• The following models were tested:
  • Ridge: R2 Score = 0.880593
  • Linear Regression: R2 Score = 0.880433
  • CatBoost Regressor: R2 Score = 0.851632
  • AdaBoost Regressor: R2 Score = 0.848774
  • XGBoost Regressor: R2 Score = 0.827797
  • Lasso: R2 Score = 0.825320
  • K-Neighbors Regressor: R2 Score = 0.784030
  • Decision Tree: R2 Score = 0.755628
• Based on the R2 scores, Ridge and Linear Regression models performed the best.
• Hyperparameter tuning is performed on the best models.
• A final VotingRegressor is created, which combines the predictions of Ridge, Linear Regression, and CatBoost models.
• This final model is saved as a pickle file.

• This pipeline converts given data into a dataframe and has various functions to load pickle files and predict the final results in Python.

• A Flask app is created with a user interface to predict the student performance scores inside a web application. ]