The present article introduces a novel model, BI.M-LSTM, which integrates the BI algorithm with the Long Short-Term Memory (LSTM) neural network to predict flight safety risks during the approach phase. Although this phase comprises only 3% of the total flight process, it is considered the most perilous stage. The proposed method involves training supervised neural networks to estimate target parameters. A standardized dataset from 2019 to 2020 was utilized, consisting of 28,813 records related to safety risk parameters such as weather conditions, aircraft configuration, flight information, speed, altitude, and air traffic. The data was summarized, cleaned, and normalized before use. Given the sequential nature of flight data and the necessity for memory retention, training was conducted using the LSTM algorithm within a Python environment. The model's mean squared error for deviations was approximately 6.38%, indicating a negligible error rate and high credibility compared to similar models. This model, enhanced with advanced tools including ETL (Extract, Transform, Load), metadata, and real-time monitoring, effectively addressed the challenges of analyzing and cleaning large-scale flight data. It successfully identified the most critical safety factor during the approach phase: control of speed and altitude during landing. This robust approach aids flight crews in managing vital safety parameters, such as preventing loss of control, maintaining appropriate aircraft speed, determining the touch-down position, and avoiding runway excursions.