The growing demand for universal food production, coupled with significant agricultural losses due to pests and crop diseases, necessitates the implementation of advanced, scalable solutions for crop health management. Machine learning (ML) has arisen as a powerful tool for automating the detection of crop diseases and pests by leveraging image-based, meteorological, and sensor data. This work presents a results-driven analysis of various ML approaches, focusing on their performance across classification and forecasting tasks. Using diverse datasets such as PlantVillage, PlantDoc, IP102, and environmental data including NDVI and temperature records, the study evaluates CNN, LSTM networks, support vector machines (SVM), and random forest algorithms. Experimental results demonstrate that CNN architectures achieve over 99% accuracy on laboratory-acquired images, while LSTM models excel in pest forecasting with approximately 92% accuracy. Transfer learning enhances model efficiency in data-scarce conditions. However, challenges persist in applying these models to field-acquired data due to variability in lighting, background, and crop species. The findings highlight the significance of dataset diversity, robust model design, and real-world validation to ensure the practical deployment of ML-based systems in agriculture.