A Machine Learning Approach to Forecasting Student Performance across Multiple Subjects

This project predicts student achievement in three essential academic domains. Machine learning methodologies were used and focused predominantly on linear regression models. These models were devised to estimate student scores based on several variables, among which were the student's gender, race/ethnicity, parental level of education, lunch type, and test preparation course completion.

Upon assessing the performance of these models, they provided a reasonable degree of accuracy. The model constructed to predict math scores exhibited a systematic bias—it consistently underestimated high scores and overestimated lower ones. This trend was gleaned from the distinctive distribution of errors. Conversely, for the reading and writing score predictions, the error distribution appeared more random, indicating a lack of consistent bias in the model's performance.

Interestingly, the variable "lunch_standard" emerged as a significant feature in these predictive models. This variable, indicating whether a student received free or reduced-price lunch, appeared to have substantial bearing on the predictions.