The aim is to predict profit that a startup will make on the basis of their R&D, Administration, Marketing Spend and State where they are based in the US.
The dataset consisted of 4 attributes on the basis of which the profit was predicted. Those were: R&D Spend, Administration Spend, Marketing Spend and the State where they are based in the U.S., namely New York, California and Florida. The R&D Spend, Administration and Marketing Spend attributes had data which was of float datatype so they didn’t require any encoding. On the other hand, the State attribute needed coding, so I label encoding it. No null values were present. Then I proceeded further and checked the variance inflation factor. On checking, the variance inflation factor for ‘R&D Spend’, ‘Administration’, ‘Marketing Spend’ and ‘State’ attributes was found to be 8.386322, 4.815916, 7.674608 and 2.382637 respectively. As all of the attributes had variance inflation factor less than 10, so I didn’t drop any of them. Moving on, I checked for outliners in the data. I used the seaborn library for plotting box plots for each and every attribute, but I found no outliners. With this, the preprocessing of the data was complete. The dataset was clean and predictions could be accurately made.
Before creating the model, it is essential to split the data for training and testing. I used the train_test_split method from model_selection in ‘sklearn’ library to perform the splitting. I split the data in the ratio of 80:20. 80% for training the model and 20% for validation. I set the random_state parameter to 0 so that I get the same results every time. This makes it easy to debug the code if required in future. With this, the splitting was efficiently done. Next, I started making the model. I initialized the Random Forest Regressor model with parameter random_state set to 0. Then, I fit the model on training set which had X_train and y_test as features. The model was successfully built and now it was the time for making the predictions on the test set and checking the accuracy. At first, I predicted results for X_test, then used the ‘score’ function for obtaining the accuracy. The score came out to be 0.9691314428968053. Hence, it means that the predictions were approximately 96.91% accurate.
I saved the model that I created in a pickle file using ‘pickle’ library. Now, in a new file, a loaded that model using ‘joblib’ library. This model will further be used to predict the ‘profit’ according to the details entered by the user. For the GUI, I imported the ‘tkinter’ library and initialized the root. Then, initialized the main canvas and wrote code to print the name of the project on the top. After this, I created labels and windows for the user to enter all the details for the attributes ‘R&D Spend’, ‘Administration’, ‘Marketing Spend’ and ‘State’. Once I had all these details ready, I called the model and added functionality for printing the results according to the entered quantities. The GUI was also ready after this and the project was complete.
Find the source here.
