32:T95f,As shown above, the data set contains **19 independent variables**, i.e. a variable whose variation does not depend on that of another. These can be classified into three groups. (a). **Demographic Information** - ```gender```: Whether the client is female or male (```Female```, ```Male```). - ```SeniorCitzen```: Whether the client is a senior citizen or not (```0```, ```1```). - ```Partner```: Whether the client has a partner or not (```Yes```, ```No```). - ```Dependants```: Whether the client has dependents or not (```Yes```, ```No```). (b). **Customer Account Information** - ```tenure```: Number of months the customer has stayed with the company(Multiple different numeric Values). - ```Contract```: Indicates the customer's current contract type (```Month-To-Month```, ```One year```, ```Two year``` ). - ```PaperlessBilling```: Whether the client has a partner or not (```Yes```, ```No```). - ```PaymentMethod```: The customer's payment method (```Electronic check```, ```Mailed check```, ```Bank transfer (automatic)```, ```Credit card (automatic)```). - ```MonthlyCharges```: The amount charged to the customer monthly (Multiple different numeric values). - ```TotalCharges```: The total amount charged to the customer(Multiple different numeric values). (c). **Services Information** - ```PhoneService```: Whether the client has a phone service or not (```Yes```, ```No```). - ```MultipleLines```: Whether the client has a partner or not (```No phone service```,```Yes```, ```No```). - ```InternetServices```: Whether the client is subscribed to Internet Service with the company (```DSL```, ```Fiber optic```, ```No```). - ```OnlineSecurity```: Whether the client has online security or not(```No internet service```,```No```, ```Yes```). - ```OnlineBackup```: Whether the client has online backup or not(```No internet service```,```No```, ```Yes```). - ```DeviceProtection```: Whether the client has device protection or not(```No internet service```,```No```, ```Yes```). - ```TechSupport```: Whether the client has tech support or not(```No internet service```,```No```, ```Yes```). - ```StreamingTV```: Whether the client has streaming TV or not(```No internet service```,```No```, ```Yes```). - ```StreamingMovies```: Whether the client has streaming movies or not(```No internet service```,```No```, ```Yes```). 33:T97c, **Features** or variables are **individual properties** e.g. **contract**, **paperlessBilling**, **totalCharges** generated from a data set and **used as input to ML models.** Usually, we represent features as numerical columns in data sets, but they can also be strings e.g **paymentMethod** in our example. **Feature Importance** expresses how much each feature **contributes** to the model prediction. It helps us determine the usefulness of a **specific feature** for a current model and prediction. This is usually represented using a score. The higher the score a value has, the more critical it is. Feature importance **scores** are beneficial in the following ways. - It is possible to determine the relationship between **independent variables** (features) and **dependent variables** (targets). - By analysing variable importance scores, we could find irrelevant features and exclude them. - Reducing the number of not-so-important variables will speed up the model or improve its performance. **Feature Importance** is commonly used as a **tool** for **ML model interpretability**. From scores, it is possible to explain why an ml model makes particular predictions and how it can manipulate features to change its predictions. **Note:** When writing this, I could not find a JavaScript library that conveniently lets me compute the **mutual information score**(mutual information measures the dependency between two variables). The way **Scikit-Learn** library does with its **metrics package**. **//TODO** Mutual information gives us a better understanding of our data and also **identifies features** that are completely **independent of our target**. Luckily we can already draw from our data visualisations **gender**, **phoneService**, and **multipleLines** will not have so much predictive power because the churn percentage for all classes(**No** or **Yes**) in all these independent variables is nearly identical. **Mutual information** allows us to use correlation criteria on nonlinear relationships, unlike the Pearson correlation that only measures linear dependence between two variables. This method of calculating feature importance **detects not only linear relationships but also nonlinear ones.** These features (**gender**, **phoneService**, and **multipleLines**) should be removed from the dataset before training as they do not provide useful information for predicting the outcome. 34:T4a4,telcoLabelEncod = { let telcoDatar = telcoData4 .clone() .replaceValues({ key: "gender", oldValue:"Female", newValue:1, }) .replaceValues({ key: "gender", oldValue:"Male", newValue:0, }) .replaceValues({ key: "partner", oldValue:"Yes", newValue:1, }) .replaceValues({ key: "partner", oldValue:"No", newValue:0, }) .replaceValues({ key: "dependents", oldValue:"Yes", newValue:1, }) .replaceValues({ key: "dependents", oldValue:"No", newValue:0, }) .replaceValues({ key: "paperlessBilling", oldValue:"Yes", newValue:1, }) .replaceValues({ key: "paperlessBilling", oldValue:"No", newValue:0, }) .replaceValues({ key: "phoneService", oldValue:"Yes", newValue:1, }) .replaceValues({ key: "phoneService", oldValue:"No", newValue:0, }) .replaceValues({ key: "churn", oldValue:"Yes", newValue:1, }) .replaceValues({ key: "churn", oldValue:"No", newValue:0, }) return telcoDatar }35:T579,model = { let [xTrain, xTest, yTrain, yTest] = spiltData; //A sequential model is any model where the outputs of one layer are the inputs to the next layer, i.e. the model topology is a simple 'stack' of layers, with no branching or skipping.... const model = tf.sequential(); //add a layer with 32 neurons model.add( tf.layers.dense({ units: 32, activation: "relu", //...This means that the first layer passed to a tf.Sequential model should have a defined input shape inputShape: [xTrain.shape[1]] }) ); //add a layer with 64 neurons model.add( tf.layers.dense({ units: 64, activation: "relu" }) ); //add a layer with 2 neurons model.add( tf.layers.dense({ units: 1, activation: "softmax" }) ); //here our neural network has two hidden layers, the inputlayer is implied from the // input shape of the first layer we created //Configures and prepares the model for training and evaluation. model.compile({ optimizer: tf.train.adam(0.001), loss: "binaryCrossentropy", metrics: ["accuracy"] }); const trainingConfig = { batchSize: 32, epochs: 32, // train model for 100 epochs shuffle: true, // shuffle the data validationSplit: 0.1, //use 10% of that data for validation } const his = await model.fit(xTrain, yTrain, trainingConfig) return his }36:[] 30:{"title":"End-to-end machine learning project: Telco customer churn","cells":[{"id":0,"value":"# End-to-end machine learning project: Telco customer churn","pinned":false,"hidden":false,"mode":"md"},{"id":11,"value":"Analyzing IBM telecommunications data [Kaggle dataset](https://www.kaggle.com/datasets/blastchar/telco-customer-churn)","pinned":false,"hidden":false,"mode":"md"},{"id":13,"value":"In order to effectively retain customers, it is crucial that **telecommunication companies** are able to predict **customer churn**. This is why large telecommunications corporations develop models to predict customer behaviour so as to take action accordingly.","pinned":false,"hidden":false,"mode":"md"},{"id":15,"value":"In this notebook, we shall build a model that predicts the likelihood of customer churn. To achieve this, customer characteristics will be analysed. These include;\n1. demographic information\n2. account information\n3. services information\n\nOur goal is to obtain a data-driven solution that will allow us to reduce churn rates and as a result, increase customer satisfaction and corporation value. \n\n","pinned":false,"hidden":false,"mode":"md"},{"id":17,"value":"#### Our Data set\n\nThe data set contains **nineteen columns (independent variables)** that indicate the **characteristics of the clients** of **SimSim**, a fictional telecommunications corporation. \n\nThe **Churn** column (response variable) indicates whether the customer departed within the last month or not. The category **No** includes the clients that did not leave the company last month, while the category **Yes** contains the clients that decided to terminate their relations with the company. Our **objective** is to **obtain the link between the customer's characteristics and the churn**.\n","pinned":false,"hidden":false,"mode":"md"},{"id":20,"value":"\n#### Project Milestones\n\nThe project consists of the following sections:\n\n1. **Data Reading**\n2. **Exploratory Data Analysis and Cleaning**\n3. **Data Visualization**\n4. **Feature Importance**\n5. **Feature Engineering**\n6. **Setting a baseline**\n7. **Splitting the data into training and testing sets**\n8. **Assessing multiple algorithms**\n9. **Algorithm selected: Gradient Boosting**\n10. **Hyperparameter tuning**\n11. **Performance of the model**\n12. **Drawing conclusions - Summary**","pinned":false,"hidden":false,"mode":"md"},{"id":9,"value":"### 1. Data Reading ","pinned":false,"hidden":false,"mode":"md"},{"id":33,"value":"To start, let's import our data.","pinned":false,"hidden":false,"mode":"md"},{"id":35,"value":"telcoData = new SimpleData()\n .loadDataFromUrl({\n url:\"https://raw.githubusercontent.com/ealecho/makailabs/main/WA_Fn-UseC_-Telco-Customer-Churn.csv\"\n })","pinned":true,"hidden":false,"mode":"ojs"},{"id":47,"value":"Let's get a description of our data and also render our data in a table to get feels of the variables we shall be interacting with.","pinned":false,"hidden":false,"mode":"md"},{"id":63,"value":"descricbeTelcoData = telcoData.clone().describe()","pinned":true,"hidden":false,"mode":"ojs"},{"id":66,"value":"showTable(descricbeTelcoData)","pinned":true,"hidden":false,"mode":"ojs"},{"id":49,"value":"showTable(telcoData)","pinned":true,"hidden":false,"mode":"ojs"},{"id":69,"value":"$32","pinned":false,"hidden":false,"mode":"md"},{"id":74,"value":"### 2. Exploratory Data Analysis and Data Cleaning\n\nHere we shall analyse the main characteristics of the dataset using **visualisation methods** and **summary statistics**. Our objective is to;\n- understand the data\n- discover patterns and anomalies\n- check assumptions before performing further evaluations.\n\n### missing values and data types\nAt the beginning of an EDA, we want to know as much information as possible, this is when the ```simpleData.checkValues()``` method comes in handy. This method tells you how many and what percentage ( count | percentage ) of the values are of a specific type.\n","pinned":false,"hidden":false,"mode":"md"},{"id":51,"value":"checkTelcoData = telcoData.clone().checkValues()","pinned":true,"hidden":false,"mode":"ojs"},{"id":53,"value":"showTable(checkTelcoData)","pinned":true,"hidden":false,"mode":"ojs"},{"id":83,"value":"As shown above, the data set contains **7043 observations** and **21 columns.** Apparently, there are no ```null``` values on the data set; however, we observe that ```seniorCitizen```, ```tenure```, ```monthlyCharges``` and ```totalCharges``` have been wrongly detected as having a data type of ```string```. These are supposed to be numeric variables, so we need to convert these columns into a **numeric** data type for further analysis. To do so, we use the ```SimpleData.valuesToInteger()``` or ```SimpleData.valuesToFloat()``` where necessary. While we are at it, let us convert the column names to [camel case convention](https://en.wikipedia.org/wiki/Camel_case) using ```SimpleData.formatAllKeys()``` method.","pinned":false,"hidden":false,"mode":"md"},{"id":95,"value":"telcoData1 = telcoData.clone().formatAllKeys()","pinned":true,"hidden":false,"mode":"ojs"},{"id":107,"value":"showTable(telcoData1)","pinned":true,"hidden":false,"mode":"ojs"},{"id":105,"value":"telcoDataCleanNum = telcoData1\n.clone()\n.valuesToInteger({key:\"seniorCitizen\"})\n.valuesToInteger({key:\"tenure\"})\n.valuesToFloat({key:\"monthlyCharges\", skipErrors: true})\n.valuesToFloat({key:\"totalCharges\", skipErrors: true}) //You can set skipErrors to true to let the original value as is.","pinned":false,"hidden":false,"mode":"ojs"},{"id":113,"value":"cleanDataCheck = telcoDataCleanNum.clone().checkValues()","pinned":false,"hidden":false,"mode":"ojs"},{"id":116,"value":"showTable(cleanDataCheck)","pinned":true,"hidden":false,"mode":"ojs"},{"id":119,"value":"In our second data check, observe ```totalCharges``` has 11 missing values. Of the 7043 observations, 7032 are numbers, while the remaining 11 are empty strings.","pinned":false,"hidden":false,"mode":"md"},{"id":124,"value":"showTable(telcoDataCleanNum)","pinned":true,"hidden":false,"mode":"ojs"},{"id":126,"value":"These 11 observations in ```totalCharges``` with missing values also have a tenure of 0. Of course, I would not expect a customer that has not spent any time on the network to have accrued a total charge; therefore, we remove those observations from the data set. We shall use the ```simpleData.excludeMissingValues()``` method to remove items (or rows in a spreadsheet way) in which one or more values are missing.","pinned":false,"hidden":false,"mode":"md"},{"id":131,"value":"telcoData2 = telcoDataCleanNum\n .clone()\n .excludeMissingValues({missingValues:[\" \"]})","pinned":false,"hidden":false,"mode":"ojs"},{"id":133,"value":"telcoData2Check = telcoData2\n .clone()\n .checkValues()","pinned":false,"hidden":false,"mode":"ojs"},{"id":135,"value":"showTable(telcoData2Check)","pinned":true,"hidden":false,"mode":"ojs"},{"id":138,"value":"showTable(telcoData2)","pinned":true,"hidden":false,"mode":"ojs"},{"id":140,"value":"### Remove customerId column\n\nThe ```customerID``` column is useless to explain whether or not the customer will churn. Therefore, let's drop it from the data set using the ```simpleData.removeKey()``` method.","pinned":false,"hidden":false,"mode":"md"},{"id":142,"value":"telcoData3 = telcoData2\n .clone()\n .removeKey({ key: \"customerId\"})","pinned":false,"hidden":false,"mode":"ojs"},{"id":146,"value":"showTable(telcoData3)","pinned":true,"hidden":false,"mode":"ojs"},{"id":148,"value":"#### Payment method denominations\n\n````paymentMethod```` has four unique values. Some payment method denominations contain in parenthesis the word automatic. These are too long to be used as tick labels in further visualizations. Let us remove the clarification in parenthesis using the SimpleData.replaceValues() method.","pinned":false,"hidden":false,"mode":"md"},{"id":155,"value":"telcoData4 = telcoData3\n .clone()\n .replaceValues({\n key: \"paymentMethod\",\n oldValue: \"automatic\",\n newValue: \"\",\n method: \"partialString\"\n })","pinned":true,"hidden":false,"mode":"ojs"},{"id":161,"value":"paymentMethods = telcoData4\n .clone()\n .getUniqueValues({key: \"paymentMethod\"})","pinned":true,"hidden":false,"mode":"ojs"},{"id":212,"value":"### 3. Data Visualization\nLet's do some analysis using visualisation.\n\n#### Response Variable\nThe following **bar plot** shows how the observations correspond to each class of the **response variable**: ``no`` and ``yes``. This is an imbalanced dataset because both classes are not equally distributed among all observations. **no** is clearly the majority class. This might affect our model later because it might lead to false negatives ie. wrongly indicating that a condition does not hold.","pinned":false,"hidden":false,"mode":"md"},{"id":247,"value":"Plot.plot({\n y:{\n grid: true,\n label:\"proportion of observations\",\n },\n marks: [\n Plot.barY(\n telcoData4.getData(),\n Plot.groupX({ y: \"proportion\", fill: \"sum\"}, { \n x: \"churn\", \n })\n )\n ],\n height: 600,\n marginLeft: 50,\n marginBottom: 50,\n width: 674\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":296,"value":"Let us use a **normalised stacked bar** to analyse the **influence of each independent categorical variable in the outcome.**\n\nA **normalised stacked bar** plot makes each column the same height, so it does not help compare total numbers; however **it is perfect for comparing how the response variable varies across all groups of an independent variable.**\n\nOn the other hand, we use **histograms** to evaluate the **influence of each independent numeric variable on the outcome.** \n","pinned":false,"hidden":false,"mode":"md"},{"id":298,"value":"### (a) Demographic Information\nLet's create a stacked percentage bar chart for each demographic attribute **(````gender````,````seniorCitizen````, ````partner````,````dependents````)**, showing the percentage of ````churn```` for each category of the attribute.","pinned":false,"hidden":false,"mode":"md"},{"id":350,"value":"Plot.plot({\n color: {\n legend: true\n },\n y: {\n grid: true,\n percent: true,\n label:\"proportion of observations by gender\",\n },\n facet: {\n data: telcoData4.getData(),\n x: \"gender\"\n },\n marks: [\n Plot.barY(telcoData4.getData(), Plot.groupZ({y: \"proportion-facet\"}, {fill: \"churn\"})),\n Plot.ruleY([0, 1])\n ],\n height: 400,\n width:400,\n marginLeft: 50,\n marginBottom: 50,\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":429,"value":"Plot.plot({\n color: {\n legend: true\n },\n y: {\n grid: true,\n percent: true,\n label:\"proportion of observations by seniorCitizen\",\n },\n facet: {\n data: telcoData4.getData(),\n x: \"seniorCitizen\"\n },\n marks: [\n Plot.barY(telcoData4.getData(), Plot.groupZ({y: \"proportion-facet\"}, {fill: \"churn\"})),\n Plot.ruleY([0, 1])\n ],\n height: 400,\n width:400,\n marginLeft: 50,\n marginBottom: 50,\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":431,"value":"Plot.plot({\n color: {\n legend: true\n },\n y: {\n grid: true,\n percent: true,\n label:\"proportion of observations by Partner\",\n },\n facet: {\n data: telcoData4.getData(),\n x: \"partner\"\n },\n marks: [\n Plot.barY(telcoData4.getData(), Plot.groupZ({y: \"proportion-facet\"}, {fill: \"churn\"})),\n Plot.ruleY([0, 1])\n ],\n height: 400,\n width:400,\n marginLeft: 50,\n marginBottom: 50,\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":434,"value":"dependentsChart = Plot.plot({\n color: {\n legend: true\n },\n y: {\n grid: true,\n percent: true,\n label:\"proportion of observations by Dependents\",\n },\n facet: {\n data: telcoData4.getData(),\n x: \"dependents\"\n },\n marks: [\n Plot.barY(telcoData4.getData(), Plot.groupZ({y: \"proportion-facet\"}, {fill: \"churn\"})),\n Plot.ruleY([0, 1])\n ],\n height: 400,\n width:400,\n marginLeft: 50,\n marginBottom: 50,\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":446,"value":"We can see that each bar is a category of the independent variable and is subdivided to show the proportion of each response class (**No** and **Yes**). We can draw the **following conclusions** from an analysis of the **demographic attributes**:\n\n- The churn rate of **senior citizens** is almost double that of **young citizens**.\n- We do not expect **gender** to have similar predictive power. A similar percentage of churn is shown when a customer is a man or woman.\n- Customers with a **partner** churn less than customers with no partner.\n\n","pinned":false,"hidden":false,"mode":"md"},{"id":473,"value":"### (b) Customer Account Information - Categorical Variables\n\nAs we did with demographic attributes, we evaluate the percentage of ````churn```` for each category of the customer attributes (````contract````, ````paperlessBilling````, ````paymentMethod````)","pinned":false,"hidden":false,"mode":"md"},{"id":478,"value":"contractsChart = Plot.plot({\n color: {\n legend: true\n },\n y: {\n grid: true,\n percent: true,\n label:\"Proportion of observations by Contract\",\n },\n facet: {\n data: telcoData4.getData(),\n x: \"contract\"\n },\n marks: [\n Plot.barY(telcoData4.getData(), Plot.groupZ({y: \"proportion-facet\"}, {fill: \"churn\"})),\n Plot.ruleY([0, 1])\n ],\n height: 400,\n width:400,\n marginLeft: 50,\n marginBottom: 50,\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":483,"value":"paperlessBillingChart = Plot.plot({\n color: {\n legend: true\n },\n y: {\n grid: true,\n percent: true,\n label:\"Proportion of observations by PaperlessBilling\",\n },\n facet: {\n data: telcoData4.getData(),\n x: \"paperlessBilling\"\n },\n marks: [\n Plot.barY(telcoData4.getData(), Plot.groupZ({y: \"proportion-facet\"}, {fill: \"churn\"})),\n Plot.ruleY([0, 1])\n ],\n height: 400,\n width:400,\n marginLeft: 50,\n marginBottom: 50,\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":488,"value":"paymentMethodChart = Plot.plot({\n color: {\n legend: true\n },\n y: {\n grid: true,\n percent: true,\n label:\"Proportion of observations by Payment method\",\n },\n facet: {\n data: telcoData4.getData(),\n x: \"paymentMethod\"\n },\n marks: [\n Plot.barY(telcoData4.getData(), Plot.groupZ({y: \"proportion-facet\"}, {fill: \"churn\"})),\n Plot.ruleY([0, 1])\n ],\n height: 400,\n width:400,\n marginLeft: 50,\n marginBottom: 50,\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":490,"value":" By analyzing **customer account attributions**, we can draw the **following conclusions.**\n - Customers with **month-to-month contracts** have higher churn rates compared to clients with **yearly contracts**\n - Customers who opted for an **electronic check** as paying method are likely to leave the company.\n - Customers subscribed to **paperless billing** churn more than those who are not subscribed.\n\n","pinned":false,"hidden":false,"mode":"md"},{"id":493,"value":"### (b) Customer Account Information - Numerical variables\n\nThe following plots show the distribution of ````tenure````, ````monthlyCharges````, ````totalCharges```` by ````churn````.For all numeric attributes, the distributions of both classes (**No** and **Yes**) are different, which suggests that all of the attributes will be useful in determining whether or not a customer churns.\n","pinned":false,"hidden":false,"mode":"md"},{"id":498,"value":"tenureChart = Plot.plot({\n color: {\n legend: true\n },\n y: {\n grid: true,\n },\n marks: [\n Plot.barY(telcoData4.getData(), Plot.binX({y: \"count\"}, {x:\"tenure\",fill: \"churn\", thresholds: 10})),\n Plot.ruleY([0, 1])\n ],\n height: 400,\n width:400,\n marginLeft: 50,\n marginBottom: 50,\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":536,"value":"monthlyChargesChart = Plot.plot({\n color: {\n legend: true\n },\n y: {\n grid: true,\n },\n marks: [\n Plot.barY(telcoData4.getData(), Plot.binX({y: \"count\"}, {x:\"monthlyCharges\",fill: \"churn\", thresholds: 10})),\n Plot.ruleY([0, 1])\n ],\n height: 400,\n width:400,\n marginLeft: 50,\n marginBottom: 50,\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":541,"value":"totalChargesChart = Plot.plot({\n color: {\n legend: true\n },\n y: {\n grid: true,\n },\n marks: [\n Plot.barY(telcoData4.getData(), Plot.binX({y: \"count\"}, {x:\"totalCharges\",fill: \"churn\", thresholds: 10})),\n Plot.ruleY([0, 1])\n ],\n height: 400,\n width:400,\n marginLeft: 50,\n marginBottom: 50,\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":552,"value":"By analysing **the histograms above** we reach the following conclusions;\n- The churn rate tends to be larger when **monthly charges** are high.\n- New customers (low **tenure**) are more likely to churn.\n- Clients with high **total charges** are less likely to leave the company. ","pinned":false,"hidden":false,"mode":"md"},{"id":557,"value":"### (c). Services Information\n\nLastly, let's evaluate the target percentage for each category of services columns with stacked bar plots.","pinned":false,"hidden":false,"mode":"md"},{"id":564,"value":"phoneServiceChart = Plot.plot({\n color: {\n legend: true\n },\n y: {\n grid: true,\n percent: true,\n label:\"Proportion of observations by Phone Service\",\n },\n facet: {\n data: telcoData4.getData(),\n x: \"phoneService\"\n },\n marks: [\n Plot.barY(telcoData4.getData(), Plot.groupZ({y: \"proportion-facet\"}, {fill: \"churn\"})),\n Plot.ruleY([0, 1])\n ],\n height: 400,\n width:400,\n marginLeft: 50,\n marginBottom: 50,\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":568,"value":"multipleLinesChart = Plot.plot({\n color: {\n legend: true\n },\n y: {\n grid: true,\n percent: true,\n label:\"Proportion of observations by Multiple Lines\",\n },\n facet: {\n data: telcoData4.getData(),\n x: \"multipleLines\"\n },\n marks: [\n Plot.barY(telcoData4.getData(), Plot.groupZ({y: \"proportion-facet\"}, {fill: \"churn\"})),\n Plot.ruleY([0, 1])\n ],\n height: 400,\n width:400,\n marginLeft: 50,\n marginBottom: 50,\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":571,"value":"internetServiceChart = Plot.plot({\n color: {\n legend: true\n },\n y: {\n grid: true,\n percent: true,\n label:\"Proportion of observations by Internet Service\",\n },\n facet: {\n data: telcoData4.getData(),\n x: \"internetService\"\n },\n marks: [\n Plot.barY(telcoData4.getData(), Plot.groupZ({y: \"proportion-facet\"}, {fill: \"churn\"})),\n Plot.ruleY([0, 1])\n ],\n height: 400,\n width:400,\n marginLeft: 50,\n marginBottom: 50,\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":574,"value":"onlineSecurityChart = Plot.plot({\n color: {\n legend: true\n },\n y: {\n grid: true,\n percent: true,\n label:\"Proportion of observations by Online Security\",\n },\n facet: {\n data: telcoData4.getData(),\n x: \"onlineSecurity\"\n },\n marks: [\n Plot.barY(telcoData4.getData(), Plot.groupZ({y: \"proportion-facet\"}, {fill: \"churn\"})),\n Plot.ruleY([0, 1])\n ],\n height: 400,\n width:400,\n marginLeft: 50,\n marginBottom: 50,\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":576,"value":"onlineBackupChart = Plot.plot({\n color: {\n legend: true\n },\n y: {\n grid: true,\n percent: true,\n label:\"Proportion of observations by Online Backup\",\n },\n facet: {\n data: telcoData4.getData(),\n x: \"onlineBackup\"\n },\n marks: [\n Plot.barY(telcoData4.getData(), Plot.groupZ({y: \"proportion-facet\"}, {fill: \"churn\"})),\n Plot.ruleY([0, 1])\n ],\n height: 400,\n width:400,\n marginLeft: 50,\n marginBottom: 50,\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":579,"value":"deviceProtectionChart = Plot.plot({\n color: {\n legend: true\n },\n y: {\n grid: true,\n percent: true,\n label:\"Proportion of observations by Device Protection\",\n },\n facet: {\n data: telcoData4.getData(),\n x: \"deviceProtection\"\n },\n marks: [\n Plot.barY(telcoData4.getData(), Plot.groupZ({y: \"proportion-facet\"}, {fill: \"churn\"})),\n Plot.ruleY([0, 1])\n ],\n height: 400,\n width:400,\n marginLeft: 50,\n marginBottom: 50,\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":583,"value":"techSupportChart = Plot.plot({\n color: {\n legend: true\n },\n y: {\n grid: true,\n percent: true,\n label:\"Proportion of observations by Tech Support\",\n },\n facet: {\n data: telcoData4.getData(),\n x: \"techSupport\"\n },\n marks: [\n Plot.barY(telcoData4.getData(), Plot.groupZ({y: \"proportion-facet\"}, {fill: \"churn\"})),\n Plot.ruleY([0, 1])\n ],\n height: 400,\n width:400,\n marginLeft: 50,\n marginBottom: 50,\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":585,"value":"streamingTVChart = Plot.plot({\n color: {\n legend: true\n },\n y: {\n grid: true,\n percent: true,\n label:\"Proportion of observations by Streaming TV\",\n },\n facet: {\n data: telcoData4.getData(),\n x: \"streamingTv\"\n },\n marks: [\n Plot.barY(telcoData4.getData(), Plot.groupZ({y: \"proportion-facet\"}, {fill: \"churn\"})),\n Plot.ruleY([0, 1])\n ],\n height: 400,\n width:400,\n marginLeft: 50,\n marginBottom: 50,\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":588,"value":"streamingMoviesChart = Plot.plot({\n color: {\n legend: true\n },\n y: {\n grid: true,\n percent: true,\n label:\"Proportion of observations by Streaming Movies\",\n },\n facet: {\n data: telcoData4.getData(),\n x: \"streamingMovies\"\n },\n marks: [\n Plot.barY(telcoData4.getData(), Plot.groupZ({y: \"proportion-facet\"}, {fill: \"churn\"})),\n Plot.ruleY([0, 1])\n ],\n height: 400,\n width:400,\n marginLeft: 50,\n marginBottom: 50,\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":595,"value":"From evaluating the **services attributes** we can draw the **following conclusions**.\n - We do not expect **phone attributes** (````phoneService```` and ````multipleLines````) to have significant predictive power. The churn percentage for all classes in both independent variables is nearly identical.\n - Clients with **online security** churn less than those without it.\n - Customers with no **tech support** tend to churn more often than those with tech support.\n\nThe above plots show us the **most relevant attributes for detecting churn**. We expect these attributes to be discriminative in our future models.","pinned":false,"hidden":false,"mode":"md"},{"id":601,"value":"## 4. Feature Importance","pinned":false,"hidden":false,"mode":"md"},{"id":603,"value":"$33","pinned":false,"hidden":false,"mode":"md"},{"id":624,"value":"## 5. Feature Engineering\n\n**Features** must be **extracted** from data **and transformed** into a format suitable for the machine learning model. This is called **Feature Engineering**. Let's transform both the numerical and categorical variables before training our model.\n- All **categorical attributes** available in the dataset should be **encoded**(transformed) into **numerical labels**\n- **Numerical columns** should be transformed into a **common scale**\n\nThis will prevent columns with large values from dominating the learning process. \n\n### Label Encoding \nCategorical values are replaced with numerical values using **label encoding**, which replaces every **category** with a **numerical label**. Let's transform the following binary variables using label encoding; \n - gender\n - partner\n - dependents\n - paperlessBilling\n - phoneService\n - churn","pinned":false,"hidden":false,"mode":"md"},{"id":610,"value":"showTable(telcoData4)","pinned":true,"hidden":false,"mode":"ojs"},{"id":612,"value":"$34","pinned":false,"hidden":false,"mode":"ojs"},{"id":614,"value":"showTable(telcoLabelEncod)","pinned":true,"hidden":false,"mode":"ojs"},{"id":628,"value":"### One-Hot Encoding\nThis creates a **new binary column for each level of categorical variable.** The new column contains zeros and ones indicating the absence or presence of the category in the data. We shall apply one-hot encoding to the following categorical variables: \n- contract\n- paymentMethod\n- multipleLines\n- internetServices\n- onlineSecurity\n- onlineBackup\n- deviceProtection\n- techSupport\n- streamingTV\n- streamingMovies\n- paperlessBilling","pinned":false,"hidden":false,"mode":"md"},{"id":662,"value":"telcoDf = new dfd.DataFrame(telcoLabelEncod.clone().getData())","pinned":true,"hidden":false,"mode":"ojs"},{"id":668,"value":"telcoOneHotEncode = {\n let one_hot_encoding_columns = ['multipleLines', 'internetService', 'onlineSecurity','onlineBackup', 'deviceProtection', 'techSupport', 'streamingTv', 'streamingMovies', 'contract', 'paymentMethod']\n let telcoDataTransformed = dfd.getDummies(telcoDf,{columns: one_hot_encoding_columns} )\n\n let telcoDataTransformedExport = dfd.toJSON(telcoDataTransformed)\n\n return new SimpleData({\n data: telcoDataTransformedExport,\n})\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":690,"value":"showTable(telcoOneHotEncode)","pinned":true,"hidden":false,"mode":"ojs"},{"id":887,"value":"telcoData4.getData()","pinned":true,"hidden":false,"mode":"ojs"},{"id":894,"value":"telcoData4.getData().map(r => r[\"gender\"])","pinned":true,"hidden":false,"mode":"ojs"},{"id":883,"value":"// nomalized \n\nnormalize = tensor => \n tf.div(\n tf.sub(tensor, tf.min(tensor)),\n tf.sub(tf.max(tensor), tf.min(tensor))\n );","pinned":true,"hidden":false,"mode":"ojs"},{"id":885,"value":"oneHot = (val, categoryCount) => \n Array.from(tf.oneHot(val, categoryCount).dataSync());","pinned":true,"hidden":false,"mode":"ojs"},{"id":881,"value":"// convert categorical features into on-hot encoded vectors\n\ntoCategorical = (data, column) => {\n\n // extract values from objects in the array using column name\n const values = data.map(r => r[column]);\n\n //keep only unique values\n const uniqueValues = new Set(values);\n\n const mapping = {};\n\n //populate our mapping object\n Array.from(uniqueValues).forEach((value, index) => {\n mapping[index] = value;\n });\n\n const encoded = values.map(v => {\n if (!v) {\n return 0; \n }\n return mapping[v];\n })\n .map(v => oneHot(v, uniqueValues.size));\n\n return encoded;\n};","pinned":true,"hidden":false,"mode":"ojs"},{"id":608,"value":"### Normalisation\n\nIt is the process of transforming **numeric columns** to a **common scale**. In machine learning, some feature values differ from others by multiple times. Features with higher values will dominate the learning process, however it doesnt meant that those variables are more important to predict the target. **Data normalization** transforms multi-scaled data to the same scale. **Normalization** causes all variables to have a similar influence on the model, improving the stability and performance of the learning algorithm.","pinned":false,"hidden":false,"mode":"md"},{"id":746,"value":"telcoDataNormalized = {\n // min-max normalization (numeric variables)\n //let min_max_columns = ['tenure', 'monthlyCharges', 'totalCharges']\n \n let newTelcoDf = new dfd.DataFrame(telcoOneHotEncode.getData())\n\n let scaler = new dfd.MinMaxScaler()\n scaler.fit(newTelcoDf)\n\n let telcoDataNormalised = dfd.toJSON(scaler.transform(newTelcoDf))\n\n return new SimpleData({\n data: telcoDataNormalised,\n})\n\n}","pinned":true,"hidden":false,"mode":"ojs"},{"id":763,"value":"showTable(telcoDataNormalized)","pinned":true,"hidden":false,"mode":"ojs"},{"id":1115,"value":"### To be completed. //TODO","pinned":false,"hidden":false,"mode":"md"},{"id":799,"value":"## Split data into training and test sets\n","pinned":false,"hidden":false,"mode":"md"},{"id":909,"value":"telcoDataTransformed = new dfd.DataFrame(telcoDataNormalized.clone().getData())","pinned":false,"hidden":false,"mode":"ojs"},{"id":912,"value":"// X_t = telcoDataTransformed.drop({columns: \"churn\"})","pinned":false,"hidden":false,"mode":"ojs"},{"id":915,"value":"// Y = telcoDataTransformed.loc({columns: [\"churn\"]})","pinned":true,"hidden":false,"mode":"ojs"},{"id":923,"value":"// telcoDataNormalized.getLength()","pinned":true,"hidden":false,"mode":"ojs"},{"id":865,"value":"\nspiltData = {\n //transform data into danfo dataframe\n let telcoDataTransformed = new dfd.DataFrame(telcoDataNormalized.getData())\n\n //select independent variables (categorical features) as a tensor \n const x = telcoDataTransformed.drop({columns: \"churn\"}).tensor\n\n const X_t = tf.tensor2d(await x.array())\n\n // select dependent variables (churn)\n const y = telcoDataTransformed.loc({columns: [\"churn\"]}).tensor\n //const X_t = telcoDataTransformed.loc({columns: [\"churn\"]}).tensor\n // return [X.tensor, Y.tensor]\n\n let testSize = 0.1;\n \n\n const splitIdx = parseInt((1- testSize) * telcoDataNormalized.getLength(), 10);\n\n const [xTrain, xTest] = tf.split(X_t, [splitIdx, telcoDataNormalized.getLength() - splitIdx]);\n const [yTrain, yTest] = tf.split(y, [splitIdx, telcoDataNormalized.getLength() - splitIdx]);\n\n return [xTrain, xTest, yTrain, yTest];\n};","pinned":true,"hidden":false,"mode":"ojs"},{"id":933,"value":"## Building a Deep Neural Network","pinned":false,"hidden":false,"mode":"md"},{"id":993,"value":"#### Define layers","pinned":false,"hidden":false,"mode":"md"},{"id":1014,"value":"$35","pinned":true,"hidden":false,"mode":"ojs"},{"id":1095,"value":"showHistory(model, ['loss', 'acc'], { height: 200 })","pinned":true,"hidden":false,"mode":"ojs"},{"id":1018,"value":"#### Compile model","pinned":false,"hidden":false,"mode":"md"},{"id":1016,"value":"\n\n// history = Error: target expected a batch of elements where each example has shape [2] (i.e.,tensor shape [*,2]) but the target received an input with 6328 examples, each with shape [40] (tensor shape [6328,40]","pinned":true,"hidden":false,"mode":"ojs"},{"id":1053,"value":"history = {\n \n const [xTrain, xTest, yTrain, yTest] = spiltData;\n const trainingConfig = {\n batchSize: 32,\n epochs: 10, // train model for 100 epochs\n shuffle: true, // shuffle the data\n validationSplit: 0.1, //use 10% of that data for validation\n }\n \n //const his = await model.fit(xTrain, yTrain, trainingConfig)\n\n}","pinned":true,"hidden":false,"mode":"ojs"},{"id":1040,"value":"lossContainer = html`

`;","pinned":true,"hidden":false,"mode":"ojs"},{"id":990,"value":"#### Train model","pinned":false,"hidden":false,"mode":"md"},{"id":7,"value":"#### Import necessary libraries\nWe use the JavaScript library [Simple data analysis](https://www.npmjs.com/package/simple-data-analysis) (SDA) by [Nael Shiab](https://www.linkedin.com/in/naelshiab/) to read and explore our data.","pinned":false,"hidden":false,"mode":"md"},{"id":5,"value":"SimpleData = {\n const sda = await import(\"https://cdn.skypack.dev/simple-data-analysis\");\n return sda.SimpleData\n};","pinned":true,"hidden":false,"mode":"ojs"},{"id":653,"value":"Import danfojs","pinned":false,"hidden":false,"mode":"md"},{"id":655,"value":"dfd = require(\"danfojs@1.0.5/lib/bundle.js\").catch(() => {\n window.dfd.Series.prototype.print = window.dfd.DataFrame.prototype.print = function () {\n return print(this);\n };\n return window.dfd;\n})","pinned":true,"hidden":false,"mode":"ojs"},{"id":871,"value":"Import tensorflow","pinned":false,"hidden":false,"mode":"md"},{"id":869,"value":"tf = require('@tensorflow/tfjs')//require(\"@tensorflow/tfjs-core\")","pinned":true,"hidden":false,"mode":"ojs"},{"id":943,"value":"tf.setBackend(\"cpu\")","pinned":true,"hidden":false,"mode":"ojs"},{"id":940,"value":"Import tensorflow visualisation library","pinned":false,"hidden":false,"mode":"md"},{"id":938,"value":"tfvis = require('@tensorflow/tfjs-vis@latest')","pinned":false,"hidden":false,"mode":"ojs"},{"id":37,"value":"#### Define some Helper Functions\nOne shall help us render our charts and the other to render tables","pinned":false,"hidden":false,"mode":"md"},{"id":973,"value":"#### showLayer(htmlOrSvg)","pinned":false,"hidden":false,"mode":"md"},{"id":39,"value":"//define some helper functions for later\nfunction showChart(htmlOrSvg) {\n return html`${htmlOrSvg}`;\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":971,"value":"#### showTable(simpleDataObject)","pinned":false,"hidden":false,"mode":"md"},{"id":41,"value":"//define some helper functions for later\nfunction showTable(sd) {\n return Inputs.table(sd.getData());\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":966,"value":"#### render(chartType, data, options)","pinned":false,"hidden":false,"mode":"md"},{"id":945,"value":"render = async (chartType, data, options) => {\n const defaultOptions = {\n width: 880,\n height: 400\n };\n\n if (!tfvis.render.hasOwnProperty(chartType)) {\n throw new Error(`Invalid chart type: ${chartType}`);\n }\n\n options = { ...defaultOptions, ...options };\n\n const elem = html`

`;\n elem.style.height = `${options.height}px`;\n\n await tfvis.render[chartType](elem, data, options);\n\n return elem;\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":963,"value":"#### showLayer(layer)","pinned":false,"hidden":false,"mode":"md"},{"id":947,"value":"showLayer = async layer => {\n const elem = html`

`;\n tfvis.show.layer(elem, layer);\n\n return elem;\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":949,"value":"showModelSummary = async model => {\n const elem = html`

`;\n tfvis.show.modelSummary(elem, model);\n\n return elem;\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":957,"value":"#### showValuesDistribution(tensor)","pinned":false,"hidden":false,"mode":"md"},{"id":951,"value":"showValuesDistribution = async tensor => {\n const elem = html`

`;\n tfvis.show.valuesDistribution(elem, tensor);\n\n return elem;\n}","pinned":true,"hidden":false,"mode":"ojs"},{"id":955,"value":"#### showHistory(history, metrics, options)","pinned":false,"hidden":false,"mode":"md"},{"id":953,"value":"showHistory = async (history, metrics, options) => {\n const defaultOptions = {\n width: 880,\n height: 400\n };\n\n options = { ...defaultOptions, ...options };\n\n const elem = html`

`;\n\n await tfvis.show.history(elem, history, metrics, options);\n\n return elem;\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":1123,"value":"### Credits","pinned":false,"hidden":false,"mode":"md"},{"id":1125,"value":"Heavily inspired by Amanda Iglesias Moreno's Medium [article](https://towardsdatascience.com/end-to-end-machine-learning-project-telco-customer-churn-90744a8df97d) on Telco customer churn prediction project in python ","pinned":false,"hidden":false,"mode":"md"}],"files":"$W36","stdlib":"1","titling":"auto"}