How to Apply Machine Learning Methods within Your Organisation | by Candice Moore
Our Data Fellowship programme is designed to equip you with the skills to manipulate data to generate insights and inform decisions.
Across the course of the apprenticeship, you will learn several machine learning methods (linear regression, logistic regression, time series forecasting etc). By the end of the programme, you will have applied some of these methods to datasets relevant to your organisation and showcased these skills in your portfolio.
By planning ahead, you can increase your chances of identifying great opportunities to apply these methods within your organisation. Not only will planning ahead give you the time needed to find the most interesting applications of these methods, it will also help you to work efficiently throughout the apprenticeship, ensuring that you have access to the right data at the right time to complete your assignments and include these in your portfolio. This resource is designed to introduce you to the machine learning methods you will come across during the apprenticeship, with examples to show how they can be applied in a business context. By understanding the aims of these methods, and some examples of how they are used, you can begin sourcing great projects so that you are ready to work on these as soon as you reach the associated modules.
Top tips for sourcing projects:
- Share the table and additional examples below with your line manager/team and discuss potential applications of these methods.
- Reach out to colleagues from other areas of the business to see if they can benefit from projects with similar aims to those listed here. Ask them to give you access to their data.
- If you are unsure which areas of the business to target, consider those departments interested in planning and resourcing (e.g. operations), forecasting (e.g. sales/marketing) or exploring data to develop interventions (e.g. HR/senior leadership).
- If you’ve carried out steps 1, 2 and 3 and are yet to identify a project, consider investigating open data sources related to your field and using these to answer questions of interest to your organisation (see here and here for open data sources).
- Remember, these types of projects are often exploratory. Conclusions may be tentative, so it’s important to manage the expectations of colleagues when discussing potential projects.
Example machine learning portfolio projects |
||||||
Portfolio Requirement |
Model Type |
Aim |
Examples |
Business Value |
Key KSB(s) |
|
|
Must do at least 1 of either linear or logistic regression |
Linear Regression |
Predicting a continuous variable from one or more other variables |
Predicting referral numbers based on previous service access, age and other factors in a health service |
Predicting resourcing needs for the service |
S11, K13, K14, S10 |
|
|
Understanding how one or more factors impact the value of a continuous variable |
Understanding which employment factors (e.g. race, gender, tenure) relate most strongly to salary |
Addressing disparities and meeting DEI targets |
S10, K13, K14 |
|||
|
Logistic Regression |
Predicting a binary (yes/no, 1/0) variable from one or more other (usually continuous) variables |
Predicting whether an email is spam or not based on the number of times certain words appear |
Reducing time spend sorting through emails |
S11, K13, K14, S10 |
||
|
Understanding how one or more variable (usually continuous) impact the value of a binary variable |
Predicting the factors most closely associated with whether or not staff members leave an organisation in the next quarter |
Exploring using these factors to reduce staff turnover |
S10, K13, K14 |
|||
|
Must do at least one of the following: time series forecasting (must be included), NLP, clustering, or decision trees. |
Time series forecasting |
Forecasting how a continuous variable will change over time, based on prior data |
Forecasting how the number of staff in an organisation will fluctuate over the next few months |
Planning onboarding resourcing |
S13, S11, K13, K14 |
|
|
NLP |
A range of methods based on analysing the words in documents to classify them or make predictions |
NLP with logistic regression to classify the sentiment of customers responding to a sexual health campaign |
Identifying themes to inform future marketing strategies |
S13, S11, K13, K14 |
||
|
Clustering |
Grouping data based on a range of features |
Grouping songs in a music app based on their speed, loudness, popularity and energy level |
Recommending new songs to users |
S13, S11, K13, K14 |
||
|
Decision tree |
Predicting the value of a categorical variable, based on one or more other variables |
Department store predicting customer purchasing based on income, family size and education level |
Forecasting profit and directing advertising spend |
S11, K13, K14, S10 |
||
Additional examples
Logistic regression
- Predicting the probability of customers defaulting on loads. Business value: to help develop interventions and reduce financial risks for lenders
- Using NLP with logistic regression to predict the sentiment of tweets based on a particular user of interest. Business value: to assess asset risk with high profile clients
- Using NLP with logistic regression to predict the sentiment of customers to a sexual health campaign in the elderly. Business value: identifying themes to inform future marketing
- Predicting the probability of a season ticket holder to renew, based on how many games they attended last season. Business value: to inform marketing strategies
- Understanding whether a software licence is likely to be breached based on the number of team members that applied, taking shared licences into consideration. Business value: to avoid licence breach
Linear Regression
- Predicting the number of new software installations for a new product. Business value: to help with resourcing and staff levels
- Predicting the length of NHS 111 calls based on patient symptoms. Business value: to help with resource planning and staff levels Identifying predictors of health outcomes in a local council. Business value: to help develop interventions and allocate funding budgets
- Predicting the length of patients stay in hospitals based on their past medical history. Business value: to help with resource planning and staff levels
- Predicting which products drive the most sales, based on where they're sold. Business value: helping to predict stock planning and recommendations
- Using conference attendees characteristics to predict how long they would attend the conference for. Business value: to inform the annual conference marketing campaign
- Predicting the number of page views on social media platforms based on the type of content/post/category etc. Business value: identifying which posts are the most effective/popular to inform content curation
- Understanding the factors that drive the time taken to close tickets raised. Business value: to determine which departments tickets took the longest time to respond to, accounting for other variables, and use this to inform resource allocation
- Predicting how much PPE will be needed based on the number of people at a site. Business value: to help with resourcing
- Predicting windspeed based on year and month. Business value: useful for planning construction projects, as windspeed has an impact on the performance of equipment such as cranes
- Predicting the strength of concrete and how long it takes to set based on other relevant features. Business value: to help with planning construction projects
- Predicting carbon value based on other features. Business value: useful for businesses aiming to quantify the environmental effects of initiatives
Time Series Forecasting
- Forecasting how revenue is likely to change over time. Business value: to inform resourcing and strategy
NLP
- Using NLP with logistic regression to predict the sentiment of tweets based on a particular user of interest. Business value: to assess asset risk with high-profile clients
- Using NLP with logistic regression to predict the sentiment of customers toward a sexual health campaign in the elderly. Business value: identifying themes to inform future marketing
Candice Moore is a Community Data Tutor at Multiverse.