Amanda

Inspiration

We were tasked with forecasting sales for the various companies based on the datasets given, taking into account a variety of factors such as number of employees (global and domestic), industry etc. We wanted to find a simple yet powerful model that could use only the crucial information to analyse and predict sales in the next quarter (?)

What it does

Our model is based mainly on the principles of linear regression, one of the basic methods under supervised learning for machines. We envisioned it to be simple enough such that even those with non-tech backgrounds would have little issue understanding and using it. Using R, we built this model to predict future sales based on a linear regression line of best fit, where global sales were plotted against various variables, such as number of employees (global), to test each variable’s impact on the sales figures.

How we built it

Firstly, we cleaned the data by removing columns/variables that were missing many values, and kept those that we wanted to use as the x-variable in the linear regression model against (global ultimate country, employees (global), ownership type, parent country, year found, 8 digit SIC code). We mainly considered these variables as we thought they were most relevant to the impact it would have on the sales figures. For instance, ownership type (public, private, non-profit etc) has a significant role to play in predicting future sales figures, as some types of organisations (eg. nonprofit) may have significantly fewer sales than the others, due to the nature of their work. The number of employees as counted globally is an indication of the scale of the company’s operations, which also gives us clues as to how they might perform in terms of sales volume. In particular, our model considered the variable Sales (Global) instead of Sales (Domestic) due to the disparity in results that our model gave. However, we acknowledge that Sales (Global) might not be as comprehensive as using Sales (Domestic) or even both Sales variables combined.

Challenges we ran into

As most of our group were beginners in the art of analysing data, we were initially unsure of where to start. More specifically, we could not decide on which programming languages (R, Python etc) to use for the project, as well as how exactly to build the model. As a result, we had to read up a lot on how we could utilise the various resources available, such as sci-kit which was relatively new to us. In addition, the data cleaning proved to be difficult as there were many columns with a significant number of missing values, stalling our progress as we had to find ways to remove them. Model selection was also a difficulty as we thought of other models, such as logistic regression, before deciding on the simpler model of linear regression.

Accomplishments that we're proud of

We managed to debug the code for the model that was previously causing issues, and as a result our code ran more smoothly. We also figured out how to better use R and Rstudio to build the models.