8.3 Analytics to Improve Decision-Making

Decision Trees

Decision trees, are commonly used to classify and predict outcomes by splitting the data based on predictor variables if data are discrete. If data are continuous, then the decision may be based on some absolute characteristic, such as whether the value is less than a certain value. For example, a bank can use a decision tree to determine whether to approve or deny a loan application. The tree might split at criteria like credit score, income level, debt-to-income ratio, and employment status. Using a decision tree standardizes the evaluation of an applicant’s criteria, automates the process of approval, ensures transparency in the approval process, and allows the bank to make data-driven decisions.

Regression

Another powerful tool in data analytics is regression, which is a statistical analysis method used to quantify the relationship between variables and to make predictions. Linear regression is one type of regression. By analyzing historical data and identifying patterns, regression models can forecast future trends and outcomes. This enables businesses to make informed decisions based on quantifiable insights. For example, a retail company can use regression to predict future sales based on several factors, such as the amount spent on advertising, seasonality, and price changes. By using linear regression, the company can model how these independent variables affect sales and make more accurate predictions. The company can also optimize their marketing budget, manage inventory more accurately, determine how price changes can impact sales, and analyze how seasonality affects sales.

Neural Networks

A neural network provides a means of machine learning by establishing a network of thousands or even millions of nodes in a weighted system of forward-moving data. Patterned loosely after biological conceptual models of how humans understand cognition, images are “trained” to provide a basis for pattern recognition. Neural networks are used in customer support chatbots to handle customer inquiries, provide recommendations, and resolve issues. Chatbots that use deep learning can enhance customer experience by understanding natural language and context and by offering personalized responses. They can also improve their accuracy in response generation and ultimately improve efficiency, leading to cost reduction and a faster response time.

Clustering

The unsupervised learning technique used to group similar data points together based on their intrinsic characteristics or attributes is called clustering. This approach is valuable for segmentation and customer profiling, allowing organizations to identify distinct groups within their target audience. By understanding these segments, businesses can tailor their marketing strategies, product offerings, and customer experiences to better meet the needs and preferences of each group. Clustering is an effective method to use in health care to improve patient care by segmenting patients based on their medical conditions, lifestyle factors, and response to treatment. The health-care organization would be able to identify groups of patients who share similar characteristics, allowing the provider to tailor treatment plans more effectively and improve outcomes.

Choosing Appropriate Models or Tools

When choosing which models to implement, organizations consider several factors. First, they evaluate the nature of the problem or decision at hand. Decision trees are often used when there are multiple decision paths, while regression analysis is suitable for predictive modeling. Clustering is employed when segmenting data is essential for personalized targeting. Second, organizations assess the availability, quality, and complexity of the data. Some models, like decision trees, provide easily interpretable results, while others, such as neural networks, may offer higher accuracy but are more challenging to interpret. Ultimately, organizations strive to select the most appropriate model that aligns with their goals, data availability, interpretability needs, and computational capabilities.

Operational, Tactical, and Strategic Decision-Making

Analytics play a vital role in helping businesses utilize different decision-making processes, including operational, tactical, and strategic decision-making. An operational decision is focused on day-to-day activities and involves optimizing processes, allocating resources, and managing immediate operational challenges. Analysts support operational decision-making by using analytics to monitor KPIs, identifying bottlenecks, and suggesting process improvements. As an example, suppose a retail store manager decides to adjust the store’s inventory levels based on daily sales data and customer demand forecasts. This decision involves managing day-to-day operations such as stocking shelves, replenishing inventory, and scheduling staff to meet immediate customer needs and maintain efficient store operations.

A tactical decision is a medium-term decision made by an organization to achieve specific objectives or goals within a defined time frame. These decisions involve resource allocation, budgeting, and setting targets. Analysts assist in tactical decision-making by conducting trend analysis, forecasting, and scenario planning. For example, a marketing manager makes a tactical decision when they launch a targeted advertising campaign for a new product line based on market research, customer segmentation analysis, and competitor benchmarking. This decision would involve developing specific marketing strategies and tactics to achieve short- to medium-term objectives, such as increasing brand awareness, expanding market share, or driving sales growth within a particular market segment.

A strategic decision is a long-term decision made by an organization to define its overall direction, goals, and competitive positioning in the market. Strategic decisions involve evaluating market trends, assessing the competitive landscape, and identifying growth opportunities. Analysts support strategic decision-making by conducting market research, competitive analysis, and trend forecasting. For example, consider a CEO of a multinational corporation who decides to enter a new international market by acquiring a competitor or forming a strategic partnership. This decision is based on comprehensive market analysis, macroeconomic trends, geopolitical factors, and long-term business goals. It involves setting overarching objectives, defining corporate strategies, and allocating resources to position the organization for sustained growth and competitive advantage in the global marketplace.

Communicating Results

Analysts use classification and prediction models to communicate results to stakeholders in a clear and understandable manner. Classification models, such as decision trees or logistic regression, are utilized to categorize data into different classes or groups. Logistic regression is a statistical modeling technique used to predict a binary or categorical outcome based on one or more independent variables. Unlike linear regression, it uses the logistic function (sigmoid curve) to model probabilities, ensuring predictions remain between zero and one. For example, logistic regression can be used to predict whether a patient has a disease based on factors like age, blood pressure, and cholesterol levels. It is widely used in classification problems, such as spam detection, customer churn prediction, or medical diagnosis.

These models help analysts communicate findings by presenting the factors or attributes that contribute to a particular classification. For example, in a marketing context, a classification model can be used to identify customer segments based on demographic or behavioral characteristics, enabling analysts to communicate the characteristics that define each segment to stakeholders.

Prediction models, such as linear regression or neural networks, are employed to make forecasts or estimate future outcomes based on historical data patterns. For example, imagine a retail company using a neural network model to predict customer purchasing behavior. By analyzing relevant data, the neural network can learn complex patterns and relationships within the data and then forecast which products customers are likely to buy in the future and anticipate changes in demand. Analysts can present the predicted values or trends to stakeholders so that they can make decisions accordingly.

Ethical and Social Considerations

Ethical concerns relating to handling data for classification and prediction models include using data for legitimate purposes, avoiding biases and discrimination, and ensuring fairness and accountability in the modeling process. Analysts must also remain objectively aware of the potential social implications of their models.

Models can inadvertently perpetuate biases or reinforce existing inequalities if the training data are biased or lack diversity. The human factor is the most important influence over bias and diversity in data. Because humans are responsible for choosing what data are fed into the algorithms and how the results will be applied, unconscious bias may enter the process if the analysts do not pay special attention to the data they use. For example, a NIST study¹¹ reported that AI facial recognition tools misidentified many people of color.

Analysts should carefully evaluate these biases to avoid negative consequences. Ensuring the ethical and unbiased use of facial recognition technology, especially in areas such as law enforcement, requires a multifaceted approach. Here are some key considerations and strategies analysts can employ:

• Ensure that the datasets used to train facial recognition algorithms are diverse and representative of the population they are meant to serve. This means including a wide range of ethnicities, ages, genders, and other relevant demographic factors in the training data.
• Implement rigorous testing procedures to detect and mitigate biases in facial recognition algorithms. This can involve analyzing the performance of the algorithm across different demographic groups and identifying any disparities in accuracy rates. Bias mitigation techniques such as algorithmic adjustments, data augmentation, and fairness-aware algorithms can help address these disparities.
• Promote transparency and accountability in the use of facial recognition technology by law enforcement agencies. This includes providing clear documentation on how the technology is used, the potential risks and limitations, and mechanisms for oversight and review by external stakeholders, including civil rights organizations and community members.

Data science professionals play an important role in effectively communicating the results of classification and prediction models to stakeholders, while simultaneously addressing ethical and social considerations to ensure responsible data handling and decision-making.