Expense report outliers are transactions that deviate significantly from the typical patterns or averages of your data. They can indicate errors, fraud, or unusual events that need further investigation. As a technical analyst (TA), you can use various methods and tools to identify and address expense report outliers for TA. In this article, you will learn how to:
Outliers are not always easy to define, as they depend on the context and the distribution of your data. A common way to identify outliers is to use the interquartile range (IQR), which is the difference between the 25th and 75th percentiles of your data. Any value that is more than 1.5 times the IQR above or below the median can be considered an outlier. However, you may need to adjust this threshold depending on your data characteristics and business objectives.
Uma forma mais eficaz de se identificar outliers é através de gráficos de distribui??o de quantidade de ocorrências por faixas fixas e de mesmo tamanho de valores considerados adequados. Aqueles valores muito mais altos do que o esperado e com uma única ocorrência ou uma quantidade ínfima de casos (menos de 0.001%), ou muitas faixas de valores vazias até que apare?a uma faixa com ocorrência de pouquíssimos casos podem ser considerados outliers.
One of the simplest ways to spot outliers is to use visual tools, such as charts and graphs, that can show the distribution and variation of your data. For example, you can use a box plot, which shows the median, quartiles, and outliers of your data as a box and whiskers. You can also use a scatter plot, which shows the relationship between two variables and any outliers that do not fit the trend. These tools can help you quickly identify any unusual or suspicious transactions in your expense reports.
Considérez également le nuage de points, qui peut être particulièrement révélateur. Il ne se limite pas à montrer les valeurs aberrantes ; il révèle également les relations entre différentes variables. Peut-être découvrirez-vous que certaines dépenses élevées sont justifiées par des projets spécifiques ou des circonstances exceptionnelles. Cette analyse bivariable peut dévoiler des insights cachés, vous permettant de prendre des décisions éclairées.
Once you have identified the outliers, you need to analyze them to understand their causes and implications. You can use various techniques, such as descriptive statistics, hypothesis testing, and clustering, to examine the outliers and compare them with the rest of your data. For example, you can use the mean, standard deviation, and z-score to measure how far an outlier is from the average. You can also use a t-test or an ANOVA to test if there is a significant difference between the outliers and the normal data. You can also use a k-means or a hierarchical clustering algorithm to group the outliers based on their similarities and differences.
Uma outra forma de se entender as causas é tentar fazer perguntas sobre a origem do dado e buscar saber por que ocorreu, ou até mesmo verificar se está correto, houve erro de digita??o ou algo parecido. O primeiro passo para se analisar outiers é validar se o valor realmente está correto ou se houve algum erro de preenchimento.
After you have analyzed the outliers, you need to decide how to address them. You can choose to keep, remove, or modify the outliers depending on your analysis results and business goals. For example, you can keep the outliers if they are valid and meaningful, such as a large expense for a special project or a rare event. You can remove the outliers if they are invalid or erroneous, such as a duplicate entry or a typo. You can modify the outliers if they are valid but extreme, such as a high expense for a long trip or a low expense for a short trip. You can use methods, such as trimming, winsorizing, or imputation, to adjust the outliers to reduce their impact on your data.
Para tratar outliers é necessário entender o impacto que este valor pode causar em sua análise antes de trata-lo. Para isto e importante ter em mente qual seu objetivo e como este resultado será utilizado. Se por exemplo um valor é muito alto, como um valor de salário para definicao de limites de crédito por exemplo, recoemnda-se trocar por um valor teto. Para tratamento em massa, poderia-se pegar os percentis 0.05 e 99.5 e qualquer valor abaixo do primeiro substituir pelo valor do percentil 0.05 e qualquer valor acima do segundo substituir pelo valor do percentil 99.5
To save time and effort, you can automate the outlier detection process using tools and scripts that can run regularly and alert you of any anomalies in your expense reports. For example, you can use Excel, Power BI, or Tableau to create dashboards and reports that can highlight the outliers and their details. You can also use Python, R, or SQL to write code that can query and filter your data and generate summary statistics and charts. You can use libraries, such as pandas, numpy, scipy, or sklearn, to perform various outlier detection methods and calculations.
Poussez l'automatisation un cran plus loin en intégrant des langages de programmation tels que Python, R, ou SQL. Ces outils offrent une flexibilité inégalée et une puissance de calcul capable de traiter des volumes massifs de données avec précision. Utilisez des bibliothèques spécialisées comme pandas pour manipuler vos données, numpy pour effectuer des calculs complexes, scipy pour appliquer des méthodes statistiques avancées, et sklearn pour implémenter des techniques d'apprentissage automatique qui peuvent reconna?tre des motifs inhabituels indiquant des valeurs aberrantes.
Finally, you need to communicate your outlier findings to your stakeholders, such as managers, clients, or auditors, who may need to take action or make decisions based on your analysis. You can use clear and concise language, visuals, and evidence to explain the outliers and their causes, effects, and solutions. You can also use recommendations, suggestions, or feedback to help your stakeholders improve their expense report policies, procedures, or practices. You can use tools, such as PowerPoint, Word, or email, to present and share your outlier findings and analysis.
O melhor a se fazer é comu ícaro momento da detec??o e n?o no final da análise para que os interessados participem da decis?o do que pode ser feito para trata-los