Author(s): Abid Ali Awan Originally published on the World’s Leading AI and Technology News and Media Company. If you are building an AI-related product or service, we invite you to consider becoming an AI sponsor. At , we help scale AI and technology startups. Let us help you unleash your technology to the masses. The main goal is to find the next ten locations similar to Saint Petersburg using unsupervised learning. Photo by Elevate on Unsplash Alcohol consumption in Russia remains among the highest in the world. According to a 2011 report by the World Health Organization, which makes it the best place to start a beverage business. People love their drinks and the company which owns a chain of stores across Russia that sell a variety of alcoholic drinks wants to invest in marketing campaigns. The company recently ran a wine promotion in Saint Petersburg that was very successful. Due to the cost to the business, it isn’t possible to run the promotion in all regions. In this project, we are going to analyze our data, fix missing values, visualize data, train the clustering model, and finally visualize our results. Data The marketing team has sourced you with historical sales volumes per capita for several different drinks types. Dataset is avaible at Alcohol Consumption in Russia (1998–2016) | Kaggle with Creative Commons — CC0 1.0 Universal License “year” — year (1998–2016) “region” — the name of a federal subject of Russia. It could be oblast, republic, krai, autonomous okrug, federal city, and a single autonomous oblast “wine” — sale of wine in liters by year per capita “beer” — sale of beer in liters by year per capita “vodka” — sale of vodka in liters by year per capita “champagne” — sale of champagne in liters by year per capita “brandy” — sale of brandy in liters by year per capita Loading Dataset We have used pandas to load the .csv dataset and its pretty small dataset containing yearly (1998-2016) alcohol consumption (beer, champagne, brandy, wine, vodka) per region (85). https://medium.com/media/a77440bc1f61c36a888793ea3551aa0f/href There are1615 sample which is logical as we have 19 years of data and 85 regions. The beer is leading the game as the mean value of beer is 51.3 liters by year per capita and the second-highest is vodka 11.81 liters by year per capita which is now even close to beer. This means people prefer beer as a go-to beverage. The beer also has the highest standard deviation which means that its demand is not stable and can fluctuate with time whereas champagne and brandy are a pretty safe bet if you want to start a low-risk business with the lowest standard deviation. https://medium.com/media/4d899888c7178b506d09531ed2734ded/href We can explore all the unique regions within the dataset. https://medium.com/media/2891042ac711b6e1c65f8282f312a400/href Correlation There is a high correlation between champagne and brandy which makes it even better. If you promote champagne there will be an increase in the sale of brandy and champagne, which makes it win-win situation. https://medium.com/media/607d086644ecc94f8b39c82bbb37e60f/href Missing Values We will be using pandas dataframe background_gradient to display several missing values and percentages. It seems like all drinks columns have missing values and the highest is brandy with 66. https://medium.com/media/f69c593dfc5155758cd076f5e6be7062/href We are going to use the fillna function and method pad to fill missing values with previous values in a column. As we can see that there are no missing values in our dataset. https://medium.com/media/d5706622a0806469d7cf12ed9bc684ab/href Geo Location For geolocation, we need to get coordinates to display stat on the map. For that we need: geopy -> Nominatim creating user agent to connect to geopy server. create lat and lon function to extract latitude and longitude using the name of the place. value count the region column, resetting index, and then renaming columns. applying both functions on to geo[‘region’] export the file as “russian_geo.csv” This process takes 5 minutes to run so we are just going to save the results in a .csv file and later merge it with our main database. Image by Author loading geo location dataset. merging it with main dataset. group by “region” and mean. sort values by “beer” descending. We are going to use df_geo dataset to plot the total alcohol consumption on the Plotly map. https://medium.com/media/b88b5f30027f9c45e49d6d31e507b63e/href Map In this section, we are going to plot the total alcohol consumption per region on the map. we have created a new column name “total” which sums up all the drinks columns. we have also created test columns that contain a caption, that will be displayed on the map. we are dividing our dataset into three categories based, first contains the top 10, the second contains 11th to 21st and third contains the rest of the regions. The ranking is based on Alcohol consumption per region. We are going to use Plotly sample code for plotting bubble maps. The top 10 and 11–21 have no paterens. They are all over the place but we can see the pattern in the rest of the categories. You can also explore the different regions by hovering your mouse over and zooming out to observe more regions. https://medium.com/media/67c2d5be5b906366e3504e1c13772488/href Alcohol Consumption Past Trend In this section, we are going to explore different types of drinks and their consumption trend over the past 19 years. As we can see beer consumption has risen with time up till 2007, then it became steady and its declining since 2011. Wine, champagne, and brandy consumption are lower than beer but they are steady. The vodka demand has increased up till 2002 and it’s declining slowly and steadily with time. The safest bet is to launch a campaign on either brandy or champagne but wine can be profitable as it has more consumption per capita and it is stable. https://medium.com/media/bef1a4fda95548c132b2cac2e18aaf9e/href Animation You can interact with the figure below and observe the change in demand by clicking on the play button. This is a simple and attractive way to present your data as a story in from of the marketing manager. It is fun too ???? https://medium.com/media/184644a4e52c97cb97ab32456c7d1fe4/href Kmean Clusters Finding cluster is a subcategory of supervised learning where there are no targets available in training data. Clustering is the task of grouping together a set of objects in a way that objects in the same cluster are more similar to each other than to objects in other clusters. K-means is vastly used for clustering in many data science applications, especially useful if you need to quickly discover insights from unlabeled data. First, we need to find how many clusters are there in our data, before that we need to create our X for that we are going to group the dataset by “region” and take average values. Then we are going to run SKlearn KMeans on a range from 1 to 9. There are 2 elbows in our line plot, we will be checking both of them. The Elbow method tells us to select the cluster when there is a significant change in inertia. We have discovered elbow at 2 and at 3. https://medium.com/media/52ac543620d33ec1e46a3e657c734900/href Let’s check the n=2 cluster. It seems quite clear, but we cannot stop here, we need to also check for n=3 in the next section. https://medium.com/media/bf10e88c4e732ae305046a9e1c28d6b6/href Now checking on n=3 cluster. I think 3 clusters are better and we are going to divide our dataset based on it. The clusters are created based on alcohol consumption. https://medium.com/media/4e2158f0b5bceac114680cfb66f482e5/href Cluster Visualization In this section we are going to compare the clusters on various columns and as you can see the best results are shown for beer V/S wine. https://medium.com/media/c7791e86c0f9e2e546d63159439ca012/href Clusters
