This paper aims to predict the churn of telecom customers, which will help us react in time and try to retain the existing users who want to switch to different networks. We will be using three different machine learning techniques for classification Support Vector Machines, K-Nearest Neighbour and Random Forest also find out the best model for classification.The data consists of information about almost six thousand users including the services they use, their demographic characteristics, the duration of the operator’s services, the amount of payment and the method of payment.In the dataset there are 20 variables, some of them which are numerical and most are categorical. There are also some missing values in the dataset. We have to do data pre- processing before implementing any model.(Data Pre-Processing) Let’s first remove the null values from the dataset. There are only 10 missing values present in total charge variable. The customers with NA values all have a tenure of 0, they are new clients who has yet to pay their bills therefore total charge value for them should be zero. We also have to drop unwanted columns like ‘gender’, ‘MultipleLines’ , ‘PhoneServices’ , ‘differences’.Exploratory Data Analysis Why the clients are more inclined to leave the company and on what factors it depends.'Phone services' were available in 91% of cases. 88 percent had a "month-to-month" contract, 82 percent had no "dependents," 78 percent had no "online security," 77 percent had no "tech support," 75 percent had "paperless billing," and 75 percent are "older citizens." 68 percent had fibre optic internet, 65 percent had no 'online backup' or 'device protection,' and 64 percent had no partner. 57 percent paid with an electronic check, 50 percent did not have'streaming TV,' were male, and did not have'streaming movies,' and 45 percent had'multiplelines.' hypotheses formulation Based on our observations, we believe that a client is more likely to depart if he has a high MonthlyCharge. This is especially true if the client is new (less than 15 months). It lacks particular services such as internet security, tech assistance, online backup, and/or device protection if the decision to quit is simple, i.e. there is no firm commitment: has a month-to-month contract, no other person involved in the decision: no dependant and/or spouse, everything can be done via the internet or over the phone: Paperless Billing and Phone Services are available.Class Imbalance It is clearly visible that there is a huge difference between the two classes (customers who stayed and the customers who left the company) one is the majority class and the other one is minority.The challenge here that we can face with such a imbalanced data is that most of the classification techniques will not consider the minority class (customers who left), and in turn show poor prediction.Here we will use one approach to address the problem SMOTE. SMOTE (Synthetic Minority Oversampling Technique) is an oversampling technique used to create synthetic samples for the minority class instead of creating copies. We will be using the from imblearn.over_sampling import SMOTE python library. The method chooses two or more comparable examples (through a distance measure) and perturbs one characteristic at a time by a random amount within the difference between the surrounding examples.The last thing we have to do is to split and scale the dataset, In splitting we will split the data into training samples and testing samples randomly and in scaling we solely normalise continuous data and leave dummy variables alone. We also apply the min-max scaler to those continuous variables, giving them the identical minimum of zero, maximum of one, and range of one. (Correlation Heatmap) Correlation heatmap is shown in the below figure it helps us to depict the relations between different variables.And also I have plotted histogram and scatter plot between variables and their relations with the target variables.We can observe that, in general, clients that desire to quit (churn = 'Yes') are new clients (low tenure 15 months, and hence low TotalCharges) with high MonthlyCharges > 65$/month. Because there is no linear relationship between tenure and TotalCharges, additional fees must be determined.( MACHINE LEARNING CLASSIFICATION TECHNIQUES) are used such as Support Vector Machine, K-Nearest Neighbor, Random Forest where Random Forest is the most perdicted accuracy model with 83.2%. random forest classification method we can get the best prediction for the customers leaving the telecom company.

How to share data with a statistician This is a guide for anyone who needs to share data with a statistician or data scientist. The target audiences I have in mind are: Collaborators who need statisticians or data scientists to analyze data for them Students or postdocs in various disciplines looking for consulting advice Junior statistics students whose job it is to collate/clean/wrangle data sets The goals of this guide are to provide some instruction on the best way to share data to avoid the most common pitfalls and sources of delay in the transition from data collection to data analysis. The Leek group works with a large number of collaborators and the number one source of variation in the speed to results is the status of the data when they arrive at the Leek group. Based on my conversations with other statisticians this is true nearly universally. My strong feeling is that statisticians should be able to handle the data in whatever state they arrive. It is important to see the raw data, understand the steps in the processing pipeline, and be able to incorporate hidden sources of variability in one's data analysis. On the other hand, for many data types, the processing steps are well documented and standardized. So the work of converting the data from raw form to directly analyzable form can be performed before calling on a statistician. This can dramatically speed the turnaround time, since the statistician doesn't have to work through all the pre-processing steps first. What you should deliver to the statistician To facilitate the most efficient and timely analysis this is the information you should pass to a statistician: The raw data. A tidy data set A code book describing each variable and its values in the tidy data set. An explicit and exact recipe you used to go from 1 -> 2,3 Let's look at each part of the data package you will transfer. The raw data It is critical that you include the rawest form of the data that you have access to. This ensures that data provenance can be maintained throughout the workflow. Here are some examples of the raw form of data: The strange binary file your measurement machine spits out The unformatted Excel file with 10 worksheets the company you contracted with sent you The complicated JSON data you got from scraping the Twitter API The hand-entered numbers you collected looking through a microscope You know the raw data are in the right format if you: Ran no software on the data Did not modify any of the data values You did not remove any data from the data set You did not summarize the data in any way If you made any modifications of the raw data it is not the raw form of the data. Reporting modified data as raw data is a very common way to slow down the analysis process, since the analyst will often have to do a forensic study of your data to figure out why the raw data looks weird. (Also imagine what would happen if new data arrived?) The tidy data set The general principles of tidy data are laid out by Hadley Wickham in this paper and this video. While both the paper and the video describe tidy data using R, the principles are more generally applicable: Each variable you measure should be in one column Each different observation of that variable should be in a different row There should be one table for each "kind" of variable If you have multiple tables, they should include a column in the table that allows them to be joined or merged While these are the hard and fast rules, there are a number of other things that will make your data set much easier to handle. First is to include a row at the top of each data table/spreadsheet that contains full row names. So if you measured age at diagnosis for patients, you would head that column with the name AgeAtDiagnosis instead of something like ADx or another abbreviation that may be hard for another person to understand. Here is an example of how this would work from genomics. Suppose that for 20 people you have collected gene expression measurements with RNA-sequencing. You have also collected demographic and clinical information about the patients including their age, treatment, and diagnosis. You would have one table/spreadsheet that contains the clinical/demographic information. It would have four columns (patient id, age, treatment, diagnosis) and 21 rows (a row with variable names, then one row for every patient). You would also have one spreadsheet for the summarized genomic data. Usually this type of data is summarized at the level of the number of counts per exon. Suppose you have 100,000 exons, then you would have a table/spreadsheet that had 21 rows (a row for gene names, and one row for each patient) and 100,001 columns (one row for patient ids and one row for each data type). If you are sharing your data with the collaborator in Excel, the tidy data should be in one Excel file per table. They should not have multiple worksheets, no macros should be applied to the data, and no columns/cells should be highlighted. Alternatively share the data in a CSV or TAB-delimited text file. (Beware however that reading CSV files into Excel can sometimes lead to non-reproducible handling of date and time variables.) The code book For almost any data set, the measurements you calculate will need to be described in more detail than you can or should sneak into the spreadsheet. The code book contains this information. At minimum it should contain: Information about the variables (including units!) in the data set not contained in the tidy data Information about the summary choices you made Information about the experimental study design you used In our genomics example, the analyst would want to know what the unit of measurement for each clinical/demographic variable is (age in years, treatment by name/dose, level of diagnosis and how heterogeneous). They would also want to know how you picked the exons you used for summarizing the genomic data (UCSC/Ensembl, etc.). They would also want to know any other information about how you did the data collection/study design. For example, are these the first 20 patients that walked into the clinic? Are they 20 highly selected patients by some characteristic like age? Are they randomized to treatments? A common format for this document is a Word file. There should be a section called "Study design" that has a thorough description of how you collected the data. There is a section called "Code book" that describes each variable and its units. How to code variables When you put variables into a spreadsheet there are several main categories you will run into depending on their data type: Continuous Ordinal Categorical Missing Censored Continuous variables are anything measured on a quantitative scale that could be any fractional number. An example would be something like weight measured in kg. Ordinal data are data that have a fixed, small (< 100) number of levels but are ordered. This could be for example survey responses where the choices are: poor, fair, good. Categorical data are data where there are multiple categories, but they aren't ordered. One example would be sex: male or female. This coding is attractive because it is self-documenting. Missing data are data that are unobserved and you don't know the mechanism. You should code missing values as NA. Censored data are data where you know the missingness mechanism on some level. Common examples are a measurement being below a detection limit or a patient being lost to follow-up. They should also be coded as NA when you don't have the data. But you should also add a new column to your tidy data called, "VariableNameCensored" which should have values of TRUE if censored and FALSE if not. In the code book you should explain why those values are missing. It is absolutely critical to report to the analyst if there is a reason you know about that some of the data are missing. You should also not impute/make up/ throw away missing observations. In general, try to avoid coding categorical or ordinal variables as numbers. When you enter the value for sex in the tidy data, it should be "male" or "female". The ordinal values in the data set should be "poor", "fair", and "good" not 1, 2 ,3. This will avoid potential mixups about which direction effects go and will help identify coding errors. Always encode every piece of information about your observations using text. For example, if you are storing data in Excel and use a form of colored text or cell background formatting to indicate information about an observation ("red variable entries were observed in experiment 1.") then this information will not be exported (and will be lost!) when the data is exported as raw text. Every piece of data should be encoded as actual text that can be exported. The instruction list/script You may have heard this before, but reproducibility is a big deal in computational science. That means, when you submit your paper, the reviewers and the rest of the world should be able to exactly replicate the analyses from raw data all the way to final results. If you are trying to be efficient, you will likely perform some summarization/data analysis steps before the data can be considered tidy. The ideal thing for you to do when performing summarization is to create a computer script (in R, Python, or something else) that takes the raw data as input and produces the tidy data you are sharing as output. You can try running your script a couple of times and see if the code produces the same output. In many cases, the person who collected the data has incentive to make it tidy for a statistician to speed the process of collaboration. They may not know how to code in a scripting language. In that case, what you should provide the statistician is something called pseudocode. It should look something like: Step 1 - take the raw file, run version 3.1.2 of summarize software with parameters a=1, b=2, c=3 Step 2 - run the software separately for each sample Step 3 - take column three of outputfile.out for each sample and that is the corresponding row in the output data set You should also include information about which system (Mac/Windows/Linux) you used the software on and whether you tried it more than once to confirm it gave the same results. Ideally, you will run this by a fellow student/labmate to confirm that they can obtain the same output file you did. What you should expect from the analyst When you turn over a properly tidied data set it dramatically decreases the workload on the statistician. So hopefully they will get back to you much sooner. But most careful statisticians will check your recipe, ask questions about steps you performed, and try to confirm that they can obtain the same tidy data that you did with, at minimum, spot checks. You should then expect from the statistician: An analysis script that performs each of the analyses (not just instructions) The exact computer code they used to run the analysis All output files/figures they generated. This is the information you will use in the supplement to establish reproducibility and precision of your results. Each of the steps in the analysis should be clearly explained and you should ask questions when you don't understand what the analyst did. It is the responsibility of both the statistician and the scientist to understand the statistical analysis. You may not be able to perform the exact analyses without the statistician's code, but you should be able to explain why the statistician performed each step to a labmate/your principal investigator. Contributors Jeff Leek - Wrote the initial version. L. Collado-Torres - Fixed typos, added links. Nick Reich - Added tips on storing data as text. Nick Horton - Minor wording suggestions.

Turn (almost) any Python command line program into a full GUI application with one line