This study proposes a machine learning-based approach for predicting forest fire occurrences one month in advance in high-risk regions of California. Multiple algorithms, including Logistic Regression (LR), Support Vector Machine (SVM), Extreme Gradient Boosting (XGB), and a Hybrid LSTM + LR model, were systematically evaluated using dynamic vegetation indices such as Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), Gross Primary Productivity (GPP), Leaf Area Index (LAI), along with their derivative components including trend and Exponential Moving Average (EMA). Static topographical and forest attributes such as slope, elevation, forest type, and forest group were also incorporated. Among the evaluated models, LR demonstrated the most consistent performance, achieving a validation Area Under the Curve (AUC) of 0.90 in Scenario F, which combined static variables with trend and EMA-based vegetation features. Experiments with varying historical windows (12, 9, and 6 months) showed that a 12-month period provided the best predictive accuracy by effectively capturing seasonal and long-term vegetation dynamics. The model was further validated using independent fire datasets from 2021 and 2024, achieving an AUC of 0.84 in 2021. However, performance declined to 0.64 in 2024, likely due to changes in satellite data products, spatial resolution, and forest phenology. The findings indicate that integrating long-term vegetation trends and EMA-based dynamic indices with static environmental variables can provide a robust and computationally efficient framework for forest fire prediction. However, future models must address data variability and environmental shifts to sustain long-term predictive performance.