Wildfires break out worldwide every year. To reduce losses, automatic, efficient, and optimal fire forecasting is necessary. In this work, we first built an extensive dataset considering a maximum number of fire-triggering factors. The proposed dataset integrates five relevant factors to wildfire events: Human factors, Land Cover, Soil Composition, Meteorological factors, and Wildfire history. To meet the need for practical fire prediction solutions, we tested several recent and promising algorithms, including Long and Short-Term Memory (LSTM), Bidirectional-LSTM (Bi-LSTM), Random Forest (RF), Logistic Regression (LR), and Artificial Neural Networks (ANN). We also experimented with different data policies, such as the data balancing policy and the model’s data feeding, separating or combining the static and time series features. Our study shows that RF methods combined with the Over-sampling balancing policy applied to the global dataset strategy gave the best results.