Objective: To predict irrigation requirements for agriculture by effectively leveraging tabular data, a domain typically dominated by traditional machine learning models.
Approach:
Pipeline Development: Designed a robust end-to-end pipeline covering data cleaning, complex feature engineering (including temperature-humidity interactions and log-scaled transformations), and category embeddings.
DL Architecture: Built a high-performing Deep Neural Network utilizing Dense layers, Batch Normalization, and Dropout to ensure stability and reduce overfitting.
Optimization: Implemented advanced training strategies such as Early Stopping and ReduceLROnPlateau to maximize accuracy and handle class imbalances via custom class weights.
Result: Achieved a top-tier score of 96.1888%, demonstrating the effectiveness of advanced Deep Learning architectures in tabular data classification tasks.
Tech Stack: Python, TensorFlow/Keras, Pandas, NumPy, Scikit-learn.
