Research Code

This repository contains the official code and results for our research paper:

Lightweight FiLM-Modulated CNN (SMCNN-MLP) vs BiGRU

The codebase contains the ablation studies, experimental results, and the final submitted manuscript demonstrating our efficiency-oriented SMCNN-MLP variant and comparing its modulation effectiveness against recurrent BiGRU sequence conditioning.

All relevant source code is located in the Paper_2_SMCNN_BiGRU/ directory, including:

• research-paper-notebook_bi_gru.ipynb / research-paper-notebook-swati-mam_latest.ipynb: The primary Jupyter Notebooks containing the data preprocessing (AFDKF, PLCWT), model architectures, training loops, and evaluation metrics.
• results_rnn_bi_gru/ & final_kaggle_cells/: The exhaustive empirical results, JSON metrics logs, generated classification maps, t-SNE plots, and confusion matrices.
• springer_manuscript.pdf: The finalized, publication-ready research manuscript incorporating all statistical tests and ablation insights.