Dataset Overview

Dataset Information

• Dataset Name: new_york_citibike
• Data Table: citybike_trips
• Data Volume: Over 58 million records
• Label: tripduration (ride duration, in seconds)
• Features: Other fields

Data Processing

• Preprocessing: Cleaning, handling missing values, converting datetime variables, feature scaling
• Data Splitting: Divide the dataset into three parts for model selection, evaluation, and testing, using month as the split criterion

Model Selection and Evaluation

• Model Choice: Linear regression model
• Evaluation Metric: Mean Squared Error (MSE)
• Model Optimization: Iterative adjustments to improve performance

Model Evaluation Results

• Model 1: trip_duration_by_stations, MSE = 111.2176
• Model 2: trip_duration_by_stations_and_day, MSE = 98.0522
• Model 3: trip_duration_by_stations_day_age, MSE = 110.8004

Conclusion

• Prediction Outcome: A total of 1,548,371 predictions were made; the predicted values differ from the actual values by less than 15 minutes
• Accuracy: In 89.6% of cases, the model predicts ride duration within 15 minutes of the actual value, and the average absolute error for predicting ride cost is 6.8 minutes

Recommended Strategy

• Pricing Model Strategy: Adopt quarterly ticket pricing and periodic payment mode
• User Story: When a user selects a bike at a start station on a specific date and specifies a destination, the model can predict ride duration and cost
• Model Performance: The model can predict ride duration and cost with accuracies of 89.6% and an average absolute error of 6.8 minutes respectively