![\"Advertisement\"](\"https:\/\/techpaathshala.com\/blog\/wp-content\/uploads\/2026\/04\/WhatsApp-Image-2026-04-20-at-11.47.35-AM.jpeg\")
<\/a><\/div>\n\n\n\nBefore the Roadmap: What Data Scientists Actually Do in Mumbai<\/h2>\n\n\n\n
The job title “data scientist” covers a wide range of actual work in practice. Understanding what Mumbai companies specifically mean by it \u2014 versus what the global discourse around data science implies \u2014 will help you build the right skills for the right market.<\/p>\n\n\n\n
In Mumbai’s FinTech and banking sector<\/strong> (Razorpay, BillDesk, HDFC, Bajaj Finance, Zerodha), data scientists primarily work on: credit risk modelling (predicting loan default probability), fraud detection (identifying anomalous transaction patterns), customer lifetime value prediction, product recommendation engines, and churn prediction. The work is heavily applied, uses well-established algorithms (logistic regression, gradient boosting, survival analysis), and requires strong SQL, Python, and domain knowledge of financial products.<\/p>\n\n\n\n In Mumbai’s e-commerce and D2C sector<\/strong> (Nykaa, Meesho, and similar), data scientists work on: demand forecasting, personalisation and recommendation systems, price optimisation, inventory management models, and A\/B test analysis at scale. The emphasis on experiment design and statistical rigour is higher here than in FinTech.<\/p>\n\n\n\n In consulting and analytics firms<\/strong> with Mumbai presence (Deloitte Analytics, EY, KPMG, boutique analytics consultancies), data scientists work on client-specific modelling projects across industries. The breadth of problems is wider, the client communication requirements are higher, and the model types vary significantly by engagement.<\/p>\n\n\n\n What this means for your roadmap:<\/strong> The skills that make you hire-ready at Mumbai’s top data science employers are not cutting-edge deep learning research skills. They are rigorous statistical foundations, strong Python for data and modelling, SQL for data access and feature engineering, and the communication ability to explain a model’s output to a business stakeholder. The roadmap that follows reflects this reality.<\/p>\n\n\n\n What it is:<\/strong> The skills without which you cannot do meaningful data science work \u2014 period. Every data scientist, regardless of how senior or specialised, has these foundations solid.<\/p>\n\n\n\n Who needs this level:<\/strong> Complete beginners with no technical background, non-CS graduates entering data science, and anyone who has been “learning data science” through YouTube videos without building these foundations deliberately and systematically.<\/p>\n\n\n\n Honest time estimate:<\/strong> 8\u201312 weeks at 1\u20131.5 hours per day for a focused learner starting from zero.<\/p>\n\n\n\n Python is the primary language of data science. Not R. Not MATLAB. Not SAS. In Mumbai’s 2026 job market, Python proficiency is the baseline technical expectation for every data science role at every level.<\/p>\n\n\n\n The Python fundamentals required for data science are not general Python mastery \u2014 you do not need to build web applications or understand async programming. You need a specific subset that enables data work.<\/p>\n\n\n\n What foundation-level Python looks like:<\/strong><\/p>\n\n\n\n Variables and data types (integers, floats, strings, booleans), lists, tuples, dictionaries, and sets \u2014 and the methods that operate on each. Conditional statements ( What foundation-level Python does not include yet:<\/strong> Pandas, NumPy, machine learning libraries \u2014 these come at the Intermediate level, built on top of this Python base.<\/p>\n\n\n\n How to build this:<\/strong> Python.org’s official tutorial, “Automate the Boring Stuff with Python” (free online), or any structured beginner Python course. The key is writing code every day, not just reading about it.<\/p>\n\n\n\n This is the skill most data science learners skip or underprioritise \u2014 and the most common reason candidates fail data science technical screens at Mumbai’s top companies.<\/p>\n\n\n\n Data science is applied mathematics. The algorithms you will use are mathematical objects. The ability to understand why a model works, why it fails, and how to improve it requires mathematical intuition \u2014 not the ability to derive proofs, but a working comfort with the concepts.<\/p>\n\n\n\n The mathematical foundation for data science:<\/strong><\/p>\n\n\n\n Linear Algebra (the essentials):<\/strong><\/p>\n\n\n\n Calculus (the essentials):<\/strong><\/p>\n\n\n\n Statistics and Probability (the essentials \u2014 and the most important for Mumbai’s market):<\/strong><\/p>\n\n\n\n What foundation-level statistics looks like in practice:<\/strong><\/p>\n\n\n\n Given a dataset of loan applicants with their approval outcomes, you can: describe the distribution of key variables, test whether approval rates differ significantly between two groups (hypothesis test), identify correlations between features, and explain what the p-value of 0.03 in a test result means in business language.<\/p>\n\n\n\n Resources:<\/strong> StatQuest with Josh Starmer (YouTube \u2014 the best free statistics resource for ML practitioners), Khan Academy for calculus, “Mathematics for Machine Learning” (free PDF from Deisenroth et al., Cambridge University Press).<\/p>\n\n\n\n SQL at the data science level goes beyond the analyst baseline. Data scientists use SQL not just to retrieve data, but to engineer features \u2014 transforming raw data in the database before it reaches Python, creating the training dataset for a model, and interrogating model outputs at scale.<\/p>\n\n\n\n Foundation-level SQL for data science:<\/strong><\/p>\n\n\n\n Everything in the data analyst SQL foundation ( This query builds an RFM (Recency, Frequency, Monetary) feature set \u2014 one of the most common feature engineering patterns in e-commerce and retail data science in Mumbai. Writing this kind of SQL is what distinguishes a data scientist from a data analyst in technical screens.<\/p>\n\n\n\n What it is:<\/strong> The Python libraries, machine learning algorithms, and model evaluation skills that constitute the working toolkit of a practising data scientist.<\/p>\n\n\n\n Who needs this level:<\/strong> Engineers and CS graduates who have Python but no ML experience. Analysts who know SQL and basic Python but have not yet built models. Anyone who has completed foundational courses but has not yet built a complete end-to-end ML project.<\/p>\n\n\n\n Honest time estimate:<\/strong> 10\u201314 weeks at 1.5 hours per day for someone who has solid foundations. Shorter for engineers with Python experience; longer for non-tech professionals building on fresh foundations.<\/p>\n\n\n\n NumPy for numerical computing:<\/strong><\/p>\n\n\n\n Pandas for data manipulation \u2014 the intermediate level:<\/strong><\/p>\n\n\n\n Visualisation for exploratory data analysis (EDA):<\/strong><\/p>\n\n\n\n The visualisation skills required at the intermediate level are not aesthetic \u2014 they are analytical. The question is whether you know which chart type reveals which kind of pattern, and whether you can interpret what you see.<\/p>\n\n\n\n Scikit-learn is the standard Python machine learning library and the tool used in the majority of production ML projects at Mumbai’s data-driven companies. The algorithms that appear most in Mumbai data science interviews and job descriptions:<\/p>\n\n\n\n Classification algorithms (for predicting categories):<\/strong><\/p>\n\n\n\n Regression algorithms (for predicting continuous values):<\/strong><\/p>\n\n\n\n Clustering algorithms (for segmentation):<\/strong><\/p>\n\n\n\n A complete classification workflow \u2014 the core skill:<\/strong><\/p>\n\n\n\n The model evaluation skills that Mumbai DS interviews test most:<\/strong><\/p>\n\n\n\n Understanding the difference between accuracy and AUC-ROC (and when each is appropriate), interpreting a confusion matrix (true positives, false positives, false negatives, true negatives), understanding precision vs. recall trade-offs (in fraud detection, false negatives are more costly; the model should be tuned accordingly), cross-validation vs. train-test split (why cross-validation gives a more reliable performance estimate), and data leakage \u2014 the most common and most serious mistake in ML pipelines (information from the test set contaminating the training process).<\/p>\n\n\n\n This is the skill that most separates experienced data scientists from those who have only completed courses. Algorithms are public knowledge \u2014 everyone can implement a Random Forest. The quality of the features you build from raw data is what determines whether your model actually works.<\/p>\n\n\n\n The feature engineering patterns that appear most in Mumbai DS work:<\/strong><\/p>\n\n\n\n For FinTech (credit, fraud, payments):<\/strong><\/p>\n\n\n\n For E-commerce (churn, LTV, recommendation):<\/strong><\/p>\n\n\n\n For HR Analytics (attrition prediction):<\/strong><\/p>\n\n\n\n The ability to look at a raw dataset and know which features to engineer \u2014 based on domain knowledge and intuition about what drives the target variable \u2014 is what makes a data scientist effective. It is learned through practice on real datasets, not through textbooks.<\/p>\n\n\n\n What it is:<\/strong> The skills that take a data scientist from “can build models” to “can build reliable, production-grade ML systems and communicate their implications to business stakeholders.”<\/p>\n\n\n\n Who needs this level:<\/strong> Analysts and intermediate practitioners aiming for senior data scientist roles (\u20b918L\u2013\u20b930L+ in Mumbai). Engineers who have built models but have not deployed them to production. Data scientists who can build models but struggle to explain them to non-technical audiences.<\/p>\n\n\n\n Honest time estimate:<\/strong> 12\u201320 weeks at 1.5\u20132 hours per day, typically overlapping with real project work rather than pure course-based learning.<\/p>\n\n\n\n XGBoost and LightGBM \u2014 the production workhorses:<\/strong><\/p>\n\n\n\n Gradient boosting models (XGBoost, LightGBM, CatBoost) dominate competition leaderboards and real-world FinTech ML applications because of their superior performance on tabular data relative to other algorithms. Understanding not just how to use them but how to tune them \u2014 learning rate, tree depth, number of estimators, regularisation parameters, early stopping \u2014 is what separates advanced practitioners from intermediate ones.<\/p>\n\n\n\n Neural Networks for structured data:<\/strong><\/p>\n\n\n\n While deep learning’s primary domain remains unstructured data (images, text, audio), neural networks are increasingly used for tabular data at Mumbai’s larger data-driven organisations. Understanding the architecture (layers, neurons, activation functions), training process (forward pass, loss calculation, backpropagation, gradient descent), regularisation techniques (dropout, batch normalisation, L1\/L2 regularisation), and implementation in TensorFlow\/Keras or PyTorch is the advanced technical skill that opens doors to roles at the intersection of data science and AI engineering.<\/p>\n\n\n\n Time Series Modelling:<\/strong><\/p>\n\n\n\n A critical specialisation for Mumbai’s FinTech and e-commerce data science teams. Demand forecasting, stock price analysis, transaction volume prediction, and customer usage patterns all involve time series data. The models that appear most in Mumbai DS JDs: ARIMA and SARIMA (classical statistical methods), Prophet (Facebook’s open-source forecasting library, widely adopted for business forecasting), and LSTM\/GRU networks for complex sequential patterns.<\/p>\n\n\n\n A model that lives only in a Jupyter notebook is not a data science product. It is an experiment. The skill of taking a model from notebook to production \u2014 where it processes real data, returns real predictions, and is monitored for performance degradation \u2014 is the advanced skill that most course-based learners never develop.<\/p>\n\n\n\n The production ML pipeline:<\/strong><\/p>\n\n\n\n Model serialisation (saving a trained model with This is what a junior deployment looks like. Understanding it, being able to build it and debug it, and knowing how to extend it to a production-grade system with authentication, rate limiting, logging, and monitoring \u2014 this is the advanced skill that justifies senior data scientist compensation.<\/p>\n\n\n\n This is the skill that the technical roadmap above does not teach \u2014 and the one that most directly determines career trajectory at the senior level.<\/p>\n\n\n\n A data scientist who can build a 94% accurate churn model but cannot explain to a product manager why 94% accuracy might still be the wrong metric \u2014 or what the model is actually measuring, and what its failure modes are \u2014 is a data scientist who will always need a technical manager between them and the business.<\/p>\n\n\n\n A data scientist who can build the same model and then walk a business stakeholder through: “here is what we are optimising for, here is the trade-off we are making between catching more churning customers and falsely flagging retained customers, here is the business action we recommend based on this model’s output, and here is how we will know if the model is degrading in production” \u2014 that data scientist is ready for a leadership track.<\/p>\n\n\n\n The communication skills that Mumbai DS roles at senior level require:<\/strong><\/p>\n\n\n\n Translating model outputs into business language (not “the AUC is 0.87” but “the model correctly identifies 8 out of 10 customers who are about to churn, with 1 in 5 of its flagged customers being a false alarm”), structuring findings as a narrative with a clear recommendation, writing a model card (a standardised document that explains a model’s purpose, training data, performance metrics, limitations, and appropriate use cases), and presenting to non-technical stakeholders without condescension or jargon.<\/p>\n\n\n\n At every level of the roadmap, the output that matters is not the course certificate. It is the work you have built and can demonstrate.<\/p>\n\n\n\n Foundation portfolio:<\/strong> A GitHub repository with Python scripts that solve real analytical problems, and SQL queries that answer real business questions on a public dataset. Evidence that you can write clean, readable code.<\/p>\n\n\n\n Intermediate portfolio:<\/strong> Two or three end-to-end ML projects on GitHub. Each project should have: a clear problem statement (the business question), a data exploration section (EDA with visualisations), a feature engineering section, at least two models compared on appropriate metrics, a clear conclusion, and a README that a non-technical person can understand. Kaggle competition placements (top 20\u201330%) are a strong supplementary signal.<\/p>\n\n\n\n Advanced portfolio:<\/strong> At least one deployed model \u2014 accessible via a public URL, demonstrating that you can move from notebook to production. Documentation that explains the model’s purpose, performance, limitations, and how its predictions should be used. Evidence of MLOps practice: reproducible training pipelines, versioned models, monitoring logic.<\/p>\n\n\n\n Skills are necessary. They are not sufficient.<\/p>\n\n\n\n The data scientists who move fastest through Mumbai’s career ladder \u2014 from fresher to senior, from senior to lead \u2014 share something that no roadmap produces: they are genuinely curious about the problems they work on. They ask questions about the business before they ask questions about the data. They read the output of their models with scepticism. They are not satisfied when the model is “good enough” if they do not understand why.<\/p>\n\n\n\n This curiosity is not teachable in a formal sense. But it is cultivatable \u2014 by working on problems you find genuinely interesting, by reading about how data science is being applied in the industries you want to work in, and by surrounding yourself with practitioners who set a standard you want to rise to.<\/p>\n\n\n\n The roadmap gives you the skills. The curiosity gives the skills somewhere meaningful to go.<\/p>\n","protected":false},"excerpt":{"rendered":" Data Scientist was called “the sexiest job of the 21st century” in 2012. In 2026, it is simply one of the most in-demand, best-compensated, and most intellectually demanding roles in Mumbai’s technology and financial ecosystem \u2014 and the path to it is clearer than it has ever been. The challenge for most people who want […]<\/p>\n","protected":false},"author":1,"featured_media":817,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"ocean_post_layout":"","ocean_both_sidebars_style":"","ocean_both_sidebars_content_width":0,"ocean_both_sidebars_sidebars_width":0,"ocean_sidebar":"","ocean_second_sidebar":"","ocean_disable_margins":"enable","ocean_add_body_class":"","ocean_shortcode_before_top_bar":"","ocean_shortcode_after_top_bar":"","ocean_shortcode_before_header":"","ocean_shortcode_after_header":"","ocean_has_shortcode":"","ocean_shortcode_after_title":"","ocean_shortcode_before_footer_widgets":"","ocean_shortcode_after_footer_widgets":"","ocean_shortcode_before_footer_bottom":"","ocean_shortcode_after_footer_bottom":"","ocean_display_top_bar":"default","ocean_display_header":"default","ocean_header_style":"","ocean_center_header_left_menu":"","ocean_custom_header_template":"","ocean_custom_logo":0,"ocean_custom_retina_logo":0,"ocean_custom_logo_max_width":0,"ocean_custom_logo_tablet_max_width":0,"ocean_custom_logo_mobile_max_width":0,"ocean_custom_logo_max_height":0,"ocean_custom_logo_tablet_max_height":0,"ocean_custom_logo_mobile_max_height":0,"ocean_header_custom_menu":"","ocean_menu_typo_font_family":"","ocean_menu_typo_font_subset":"","ocean_menu_typo_font_size":0,"ocean_menu_typo_font_size_tablet":0,"ocean_menu_typo_font_size_mobile":0,"ocean_menu_typo_font_size_unit":"px","ocean_menu_typo_font_weight":"","ocean_menu_typo_font_weight_tablet":"","ocean_menu_typo_font_weight_mobile":"","ocean_menu_typo_transform":"","ocean_menu_typo_transform_tablet":"","ocean_menu_typo_transform_mobile":"","ocean_menu_typo_line_height":0,"ocean_menu_typo_line_height_tablet":0,"ocean_menu_typo_line_height_mobile":0,"ocean_menu_typo_line_height_unit":"","ocean_menu_typo_spacing":0,"ocean_menu_typo_spacing_tablet":0,"ocean_menu_typo_spacing_mobile":0,"ocean_menu_typo_spacing_unit":"","ocean_menu_link_color":"","ocean_menu_link_color_hover":"","ocean_menu_link_color_active":"","ocean_menu_link_background":"","ocean_menu_link_hover_background":"","ocean_menu_link_active_background":"","ocean_menu_social_links_bg":"","ocean_menu_social_hover_links_bg":"","ocean_menu_social_links_color":"","ocean_menu_social_hover_links_color":"","ocean_disable_title":"default","ocean_disable_heading":"default","ocean_post_title":"","ocean_post_subheading":"","ocean_post_title_style":"","ocean_post_title_background_color":"","ocean_post_title_background":0,"ocean_post_title_bg_image_position":"","ocean_post_title_bg_image_attachment":"","ocean_post_title_bg_image_repeat":"","ocean_post_title_bg_image_size":"","ocean_post_title_height":0,"ocean_post_title_bg_overlay":0.5,"ocean_post_title_bg_overlay_color":"","ocean_disable_breadcrumbs":"default","ocean_breadcrumbs_color":"","ocean_breadcrumbs_separator_color":"","ocean_breadcrumbs_links_color":"","ocean_breadcrumbs_links_hover_color":"","ocean_display_footer_widgets":"default","ocean_display_footer_bottom":"default","ocean_custom_footer_template":"","ocean_post_oembed":"","ocean_post_self_hosted_media":"","ocean_post_video_embed":"","ocean_link_format":"","ocean_link_format_target":"self","ocean_quote_format":"","ocean_quote_format_link":"post","ocean_gallery_link_images":"on","ocean_gallery_id":[],"footnotes":""},"categories":[71],"tags":[],"class_list":["post-797","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-data-science","entry","has-media"],"acf":[],"_links":{"self":[{"href":"https:\/\/techpaathshala.com\/blog\/wp-json\/wp\/v2\/posts\/797","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/techpaathshala.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/techpaathshala.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/techpaathshala.com\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/techpaathshala.com\/blog\/wp-json\/wp\/v2\/comments?post=797"}],"version-history":[{"count":2,"href":"https:\/\/techpaathshala.com\/blog\/wp-json\/wp\/v2\/posts\/797\/revisions"}],"predecessor-version":[{"id":914,"href":"https:\/\/techpaathshala.com\/blog\/wp-json\/wp\/v2\/posts\/797\/revisions\/914"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/techpaathshala.com\/blog\/wp-json\/wp\/v2\/media\/817"}],"wp:attachment":[{"href":"https:\/\/techpaathshala.com\/blog\/wp-json\/wp\/v2\/media?parent=797"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/techpaathshala.com\/blog\/wp-json\/wp\/v2\/categories?post=797"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/techpaathshala.com\/blog\/wp-json\/wp\/v2\/tags?post=797"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
\n\n\n\nFoundation Level: The Non-Negotiable Starting Point<\/h2>\n\n\n\n
\n\n\n\nFoundation Skill 1: Python Programming Fundamentals<\/h3>\n\n\n\n
if<\/code>, elif<\/code>, else<\/code>). Loops (for<\/code>, while<\/code>) and list comprehensions. Functions \u2014 defining them, passing arguments, returning values, understanding scope. File I\/O \u2014 reading and writing CSV and text files. Error handling with try<\/code>\/except<\/code>. Installing and importing libraries with pip<\/code> and import<\/code>.<\/p>\n\n\n\n# The kind of Python a foundation-level learner should be comfortable writing\n\ndef analyse_sales(filepath):\n \"\"\"Load a CSV file and return basic sales metrics.\"\"\"\n sales = []\n\n try:\n with open(filepath, 'r') as f:\n next(f) # skip header\n for line in f:\n parts = line.strip().split(',')\n amount = float(parts[2])\n city = parts[3]\n sales.append({'amount': amount, 'city': city})\n except FileNotFoundError:\n print(f\"File not found: {filepath}\")\n return None\n\n total = sum(item['amount'] for item in sales)\n avg = total \/ len(sales) if sales else 0\n cities = list(set(item['city'] for item in sales))\n\n return {\n 'total_sales': total,\n 'avg_sale': avg,\n 'num_transactions': len(sales),\n 'cities': cities\n }\n\nresult = analyse_sales('mumbai_sales.csv')\nprint(result)\n<\/code><\/pre>\n\n\n\n
\n\n\n\nFoundation Skill 2: Mathematics and Statistics<\/h3>\n\n\n\n
\n
\n
\n
\n\n\n\nFoundation Skill 3: SQL for Data Science<\/h3>\n\n\n\n
SELECT<\/code>, WHERE<\/code>, JOIN<\/code>, GROUP BY<\/code>, CTEs) plus: window functions (ROW_NUMBER<\/code>, LAG<\/code>, LEAD<\/code>, NTILE<\/code>, PERCENT_RANK<\/code> \u2014 used extensively for feature engineering), advanced aggregation (ROLLUP<\/code>, CUBE<\/code> for multi-level summaries), self-joins (joining a table to itself \u2014 used for time-based feature engineering), and date manipulation for creating time-based features (days since last transaction, rolling 30-day averages).<\/p>\n\n\n\n-- Feature engineering example: customer recency, frequency, monetary value (RFM)\n-- This is the kind of SQL a data scientist writes before model training\n\nWITH customer_rfm AS (\n SELECT\n customer_id,\n -- Recency: days since last order\n DATEDIFF(CURRENT_DATE, MAX(order_date)) AS days_since_last_order,\n -- Frequency: number of orders in last 12 months\n COUNT(DISTINCT order_id) AS order_count_12m,\n -- Monetary: total spend in last 12 months\n SUM(order_amount) AS total_spend_12m,\n -- Average order value\n AVG(order_amount) AS avg_order_value,\n -- Days between first and last order (customer tenure)\n DATEDIFF(MAX(order_date), MIN(order_date)) AS customer_tenure_days\n FROM orders\n WHERE order_date >= DATE_SUB(CURRENT_DATE, INTERVAL 12 MONTH)\n GROUP BY customer_id\n),\nrfm_scored AS (\n SELECT\n customer_id,\n days_since_last_order,\n order_count_12m,\n total_spend_12m,\n avg_order_value,\n customer_tenure_days,\n -- Quintile scoring for each RFM dimension\n NTILE(5) OVER (ORDER BY days_since_last_order ASC) AS recency_score,\n NTILE(5) OVER (ORDER BY order_count_12m DESC) AS frequency_score,\n NTILE(5) OVER (ORDER BY total_spend_12m DESC) AS monetary_score\n FROM customer_rfm\n)\nSELECT\n customer_id,\n recency_score,\n frequency_score,\n monetary_score,\n (recency_score + frequency_score + monetary_score) AS rfm_total_score\nFROM rfm_scored\nORDER BY rfm_total_score DESC;\n<\/code><\/pre>\n\n\n\n
\n\n\n\nIntermediate Level: The Core Data Science Toolkit<\/h2>\n\n\n\n
\n\n\n\nIntermediate Skill 1: Python Data Stack \u2014 NumPy, Pandas, and Visualisation<\/h3>\n\n\n\n
import numpy as np\n\n# The operations data scientists use NumPy for most\narr = np.array([14, 22, 8, 35, 17, 44, 9, 28])\n\nprint(f\"Mean: {np.mean(arr):.2f}\")\nprint(f\"Std Dev: {np.std(arr):.2f}\")\nprint(f\"Median: {np.median(arr):.2f}\")\nprint(f\"25th percentile: {np.percentile(arr, 25):.2f}\")\nprint(f\"75th percentile: {np.percentile(arr, 75):.2f}\")\n\n# Boolean masking \u2014 filtering arrays by condition\nabove_mean = arr[arr > np.mean(arr)]\nprint(f\"Values above mean: {above_mean}\")\n\n# Reshaping \u2014 critical for ML input preparation\nmatrix = arr.reshape(2, 4)\nprint(f\"Reshaped to 2x4:\\n{matrix}\")\n<\/code><\/pre>\n\n\n\nimport pandas as pd\n\ndf = pd.read_csv('mumbai_customers.csv')\n\n# Missing value strategy\nprint(df.isnull().sum())\ndf['age'].fillna(df['age'].median(), inplace=True)\ndf.dropna(subset=['customer_id', 'city'], inplace=True)\n\n# Feature creation\ndf['is_mumbai'] = (df['city'] == 'Mumbai').astype(int)\ndf['high_value'] = (df['total_spend'] > df['total_spend'].quantile(0.75)).astype(int)\ndf['log_spend'] = np.log1p(df['total_spend']) # log transform for skewed distribution\n\n# Merging datasets\ntransactions = pd.read_csv('transactions.csv')\nmerged = df.merge(\n transactions.groupby('customer_id').agg(\n num_transactions=('order_id', 'count'),\n avg_transaction=('amount', 'mean'),\n last_transaction=('date', 'max')\n ).reset_index(),\n on='customer_id',\n how='left'\n)\n\n# Time-based feature engineering\nmerged['last_transaction'] = pd.to_datetime(merged['last_transaction'])\nmerged['days_since_last_txn'] = (pd.Timestamp.today() - merged['last_transaction']).dt.days\n<\/code><\/pre>\n\n\n\nimport matplotlib.pyplot as plt\nimport seaborn as sns\n\nfig, axes = plt.subplots(2, 2, figsize=(14, 10))\n\n# Distribution of key variable\nsns.histplot(df['total_spend'], bins=50, kde=True, ax=axes[0, 0])\naxes[0, 0].set_title('Distribution of Customer Spend (note: right skew)')\n\n# Correlation heatmap \u2014 identify multicollinearity before modelling\nnumeric_cols = df.select_dtypes(include=[np.number]).columns\ncorrelation_matrix = df[numeric_cols].corr()\nsns.heatmap(correlation_matrix, annot=True, fmt='.2f', cmap='coolwarm', ax=axes[0, 1])\naxes[0, 1].set_title('Feature Correlation Matrix')\n\n# Churn rate by segment\nchurn_by_city = df.groupby('city')['churned'].mean().sort_values(ascending=False)\nchurn_by_city.plot(kind='bar', ax=axes[1, 0])\naxes[1, 0].set_title('Churn Rate by City')\naxes[1, 0].set_ylabel('Churn Rate')\n\n# Spend distribution by churn status\ndf.boxplot(column='total_spend', by='churned', ax=axes[1, 1])\naxes[1, 1].set_title('Spend Distribution by Churn Status')\n\nplt.tight_layout()\nplt.savefig('eda_dashboard.png', dpi=150, bbox_inches='tight')\nplt.show()\n<\/code><\/pre>\n\n\n\n
\n\n\n\nIntermediate Skill 2: Machine Learning with Scikit-learn<\/h3>\n\n\n\n
\n
\n
\n
import pandas as pd\nimport numpy as np\nfrom sklearn.model_selection import train_test_split, cross_val_score, StratifiedKFold\nfrom sklearn.preprocessing import StandardScaler, LabelEncoder\nfrom sklearn.ensemble import RandomForestClassifier, GradientBoostingClassifier\nfrom sklearn.linear_model import LogisticRegression\nfrom sklearn.metrics import (classification_report, confusion_matrix,\n roc_auc_score, roc_curve)\nfrom sklearn.pipeline import Pipeline\nimport matplotlib.pyplot as plt\nimport warnings\nwarnings.filterwarnings('ignore')\n\n# --- DATA PREPARATION ---\ndf = pd.read_csv('customer_churn_features.csv')\n\n# Encode categorical variables\nle = LabelEncoder()\ndf['city_encoded'] = le.fit_transform(df['city'])\ndf['segment_encoded'] = le.fit_transform(df['segment'])\n\n# Define features and target\nfeature_cols = [\n 'age', 'tenure_months', 'total_spend_12m', 'avg_order_value',\n 'num_transactions', 'days_since_last_order', 'city_encoded',\n 'segment_encoded', 'log_spend'\n]\n\nX = df[feature_cols]\ny = df['churned']\n\nprint(f\"Class distribution:\\n{y.value_counts(normalize=True).round(3)}\")\n\n# --- TRAIN-TEST SPLIT ---\n# Stratified split preserves class proportions in both sets\nX_train, X_test, y_train, y_test = train_test_split(\n X, y, test_size=0.2, random_state=42, stratify=y\n)\n\n# --- MODEL TRAINING WITH PIPELINE ---\n# Pipeline ensures scaler is fit only on training data (prevents data leakage)\npipeline = Pipeline([\n ('scaler', StandardScaler()),\n ('model', RandomForestClassifier(\n n_estimators=200,\n max_depth=8,\n min_samples_leaf=10,\n class_weight='balanced', # handles class imbalance\n random_state=42\n ))\n])\n\npipeline.fit(X_train, y_train)\n\n# --- EVALUATION ---\ny_pred = pipeline.predict(X_test)\ny_prob = pipeline.predict_proba(X_test)[:, 1]\n\nprint(\"\\nClassification Report:\")\nprint(classification_report(y_test, y_pred))\nprint(f\"AUC-ROC Score: {roc_auc_score(y_test, y_prob):.4f}\")\n\n# Cross-validation for robust performance estimate\ncv = StratifiedKFold(n_splits=5, shuffle=True, random_state=42)\ncv_scores = cross_val_score(pipeline, X, y, cv=cv, scoring='roc_auc')\nprint(f\"\\n5-Fold CV AUC: {cv_scores.mean():.4f} (+\/- {cv_scores.std():.4f})\")\n\n# --- FEATURE IMPORTANCE ---\nrf_model = pipeline.named_steps['model']\nimportance_df = pd.DataFrame({\n 'feature': feature_cols,\n 'importance': rf_model.feature_importances_\n}).sort_values('importance', ascending=False)\n\nprint(\"\\nTop 5 Most Important Features:\")\nprint(importance_df.head())\n<\/code><\/pre>\n\n\n\n
\n\n\n\nIntermediate Skill 3: Feature Engineering<\/h3>\n\n\n\n
\n
\n
\n
\n\n\n\nAdvanced Level: The Specialisation That Commands Senior Salaries<\/h2>\n\n\n\n
\n\n\n\nAdvanced Skill 1: Advanced Algorithms and Ensemble Methods<\/h3>\n\n\n\n
\n\n\n\nAdvanced Skill 2: Model Deployment and MLOps Fundamentals<\/h3>\n\n\n\n
pickle<\/code> or joblib<\/code> so it can be loaded without retraining), building a REST API to serve predictions (Flask or FastAPI in Python), containerising the model with Docker, deploying to a cloud platform (AWS SageMaker, Google Vertex AI, Azure ML), and setting up monitoring for model performance in production (detecting data drift, monitoring prediction distribution, alerting on performance degradation).<\/p>\n\n\n\n# A minimal FastAPI model serving endpoint\nfrom fastapi import FastAPI\nfrom pydantic import BaseModel\nimport joblib\nimport pandas as pd\nimport numpy as np\n\napp = FastAPI(title=\"Churn Prediction API\")\n\n# Load trained model at startup\nmodel = joblib.load('churn_model_pipeline.pkl')\n\nclass CustomerFeatures(BaseModel):\n age: float\n tenure_months: float\n total_spend_12m: float\n avg_order_value: float\n num_transactions: int\n days_since_last_order: int\n city_encoded: int\n segment_encoded: int\n log_spend: float\n\nclass PredictionResponse(BaseModel):\n customer_id: str\n churn_probability: float\n churn_prediction: bool\n risk_segment: str\n\n@app.post(\"\/predict\", response_model=PredictionResponse)\nasync def predict_churn(customer_id: str, features: CustomerFeatures):\n feature_df = pd.DataFrame([features.dict()])\n churn_prob = model.predict_proba(feature_df)[0, 1]\n churn_pred = churn_prob > 0.5\n\n risk_segment = (\n \"High Risk\" if churn_prob > 0.7 else\n \"Medium Risk\" if churn_prob > 0.4 else\n \"Low Risk\"\n )\n\n return PredictionResponse(\n customer_id=customer_id,\n churn_probability=round(float(churn_prob), 4),\n churn_prediction=bool(churn_pred),\n risk_segment=risk_segment\n )\n<\/code><\/pre>\n\n\n\n
\n\n\n\nAdvanced Skill 3: Communication and Business Translation<\/h3>\n\n\n\n
\n\n\n\nThe Portfolio That Gets You Hired in Mumbai<\/h2>\n\n\n\n
Experience Level<\/th> Skill Level<\/th> Typical Salary Range (\u20b9 LPA)<\/th> Key Skills<\/th> Hiring Demand in Mumbai<\/th><\/tr><\/thead> 0\u20131 Years<\/td> Beginner<\/td> \u20b94 \u2013 \u20b98 LPA<\/td> Python, Excel, Basic Statistics, SQL<\/td> Moderate (freshers, internships, trainees)<\/td><\/tr> 1\u20133 Years<\/td> Junior<\/td> \u20b96 \u2013 \u20b912 LPA<\/td> Python, Pandas, Data Visualization, SQL, ML Basics<\/td> High (startups & analytics firms)<\/td><\/tr> 3\u20135 Years<\/td> Mid-Level<\/td> \u20b910 \u2013 \u20b920 LPA<\/td> Machine Learning, Feature Engineering, APIs, Cloud Basics<\/td> Very High (fintech, e-commerce, SaaS)<\/td><\/tr> 5\u20138 Years<\/td> Senior<\/td> \u20b918 \u2013 \u20b935 LPA<\/td> Deep Learning, NLP, Big Data (Spark), MLOps<\/td> Very High (product companies, AI teams)<\/td><\/tr> 8+ Years<\/td> Lead \/ Principal<\/td> \u20b930 \u2013 \u20b960+ LPA<\/td> AI Strategy, System Design, Team Leadership, Advanced AI<\/td> Critical demand (leadership roles)<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
\n\n\n\nThe Most Important Thing This Roadmap Cannot Tell You<\/h2>\n\n\n\n