# Regression techniques to predict hourly transactions?

First of all, I self study Data Science from online courses, so please bear with me.

I have a regression problem with a dataset that similar to Bay Area Bike Share. The goal of this project is to predict the number of "transaction rate" in each hour. Transaction rate is the difference of number of start trips and number of end trips.

For example, let's say we have start trips = 3 and end trips = 5 in a particular hour at a particular station, then the transaction rate will be 5-3 = 2.

There will be no transaction rate if start trips == end trips or if there are no transaction at particular hour. The transaction rate can also be negative if start trips > end trips.

I also have weather data on each day & each station, so the final features will be the weather condition & net rate of each hour (the hour will be a feature, along with month and day). Because the weather data is available at each day, all 24 hours in a particular day will have same weather data.

Long story short, I tried to do the linear regression using these features and the result is inaccurate. The R^2 score is just around 0.2 and the MAE is 0.5. I do compare the prediction result and confirmed that the predictions are plain wrong.

I tried to check the univariate (Pearson) correlation of the features with target variable, and found out that even the strongest correlated feature only have 0.1/0.2 correlation.

Does there exist a better way to predict the regression on hourly rate?

Note : I will update the statistical details of "net rate" later, I don't bring the data with me.