The initial phase of the project involved training a neural network model from scratch using NumPy for SAT score prediction. The results and findings from this initial exploration phase are detailed below.
-
Training Process: The initial model was trained using a dataset in the original csv file that included SAT scores and related input features in
data, with SAT Score being a primary focus. -
Promising Potential: The results of this initial training phase were highly promising. The model exhibited significant potential, achieving an accuracy of approximately 65% after 30 epochs of training. This marked the foundation upon which subsequent optimizations and enhancements were built.
-
Regression Focus: The primary objective of this initial model was to establish a robust regression model capable of accurately predicting SAT scores based on various input factors, with GPA being a key predictor.
The success of this initial model has paved the way for further refinements and enhancements. Future steps in this project will involve:
-
Hyperparameter Tuning: Fine-tuning the model's hyperparameters to optimize its performance, potentially leading to higher accuracy and faster convergence.
-
Data Preprocessing: Refining data preprocessing techniques to handle missing values, scale input features, and conduct feature engineering as needed.
-
Loss Function Evaluation: Experimenting with different loss functions to identify the one that best suits the regression task.
-
Visualization: Continued monitoring of the model's training progress and outcomes through visualizations to gain insights into its learning process.
-
Extension to Classification: The project will be extended to include a classification aspect to predict college admission outcomes based on the academic profile.
This initial exploration phase has set the stage for a more comprehensive and accurate SAT score prediction model, with the ultimate aim of providing valuable insights and predictions to students, educators, and institutions.