This text presents an analysis of data aimed at improving the model for predicting taxi ride durations in New York City. The dataset used includes timestamps for trip start times, geographic coordinates, the number of passengers, and other variables, with the target variable being trip duration, which shows an asymmetric distribution with a long right tail. RMSLE is used as the evaluation metric instead of RMSE due to the greater importance of relative errors in predicting trip duration. The analysis of temporal and geographic data revealed significant variations in taxi demand depending on the day of the week, time of day, weather conditions, and holidays. Key points of activity, such as train stations and schools, were identified, and taxi speeds during different periods were examined. The model improvement process also involved handling outliers and combining rare categorical feature categories to prevent overfitting. The conclusion emphasizes the need to consider temporal and geographic features to enhance the accuracy and interpretability of the taxi ride duration prediction model.