Predicting visitor numbers to Blackpool

• We can see Sunday had a big effect in predicting significantly less cars on the road.
• The month of the year is also important with April (Easter) predicting more visitors and December predicting less visitors.
• More data is needed with other features to improve the prediction.

Following on from a previous post I have been exploring the features that go into predicting visitor numbers to Blackpool. Using statistical modelling and machine learning models I explored the data from motorway M55 traffic over several years. The traffic will be a mix of local commuters and visitors to the town. The fluctuation should give us an idea of trends such as day of the week and time of the year. Summer brings more visitors to the town and weekends bring more visitors.

After fitting the models we can see what features the model found to be most important to predicting the car numbers. Data from previous years specifically 2013 and 2014 were obviously important factors. What happened in the last few years is likely to repeat in future years. We can see weekday6 which is Sunday had a big effect in predicting significantly less cars on the road. This is likely because of the reduction in locals commuting to work. A peak on Friday and Saturday is likely to be overnight visitors who are arriving into the town for the weekend. Other factors include the weather with rain and wind predicting less visitors and higher temperatures predicting more visitors. The month of the year is also important with April (Easter) predicting more visitors and December predicting less visitors. September also predicts less visitors so more exploration into why that is would be a good idea.

This is the result of another machine learning model I fitted. This only shows the importance of each feature and not if it reduces or increases the prediction of cars. Friday and Sunday are important features along with the 2017 values and weather features. Gtrend is google trends and this looks at the mention of Blackpool for that specific day of the year. If more people are searching for Blackpool then it is a predictor of increased visitor numbers.

This scatter plot shows predicted values vs real values. The correlation is clearly close with quite a spread of values showing we can predict visitor numbers for any specific date but the variance is quite wide so the certainty of our prediction is not too strong. There are clearly other factors which go into estimating visitor numbers. The machine learning model is 60% accurate which means 60% of the information about visitor numbers is actually in the data. More data is needed with other features to improve the prediction.