# UTS — Pembelajaran Mesin (Reg B) (2022) > Midterm Exam · Universitas Gadjah Mada > **Note:** This source is a **Moodle quiz attempt export** (22 multiple-choice questions, 1 mark each), not a printed exam paper. The quiz pages carry no university letterhead, no date and no lecturer field — those values are therefore not recoverable from the source. | Field | Value | |---|---| | **Course** | Pembelajaran Mesin | | **Exam type** | Ujian Tengah Semester (UTS) | | **Class** | Reg B | | **Year** | 2022 | | **Format** | Online quiz (Moodle), multiple choice | | **Questions** | 22 (Q1–Q23 with **Q12 absent** in source) | | **Marking** | 1.00 mark per question | **Institution:** Program Studi S1 Ilmu Komputer, Fakultas Matematika dan Ilmu Pengetahuan Alam, Universitas Gadjah Mada **Transcription scope:** All 22 questions in the source PDF. Each question is reproduced with its option list, followed by the **option that was selected in the attempt** (`Recorded selection`), as indicated by the filled radio button in the export. --- ## Question 1 Logistic regression is a method for - a. forecasting - b. classification - c. regression - d. supervised learning **Recorded selection:** b — classification --- ## Question 2 Higher the P-value, - a. lower the F-score - b. higher the significance of statistics - c. higher the F-score - d. lower the significance of statistics **Recorded selection:** d — lower the significance of statistics --- ## Question 3 There are some methods to detect outlier data, except - a. distance-base clustering *(sic — source reads "distance-base")* - b. visualization - c. Tukey's fences - d. 3-sigma rule **Recorded selection:** b — visualization --- ## Question 4 Different with supervised learning, reinforcement learning gives an output - a. cluster - b. label - c. reward - d. action **Recorded selection:** c — reward --- ## Question 5 The data type of a number of covid-19 new cases each days is - a. structured data - b. supervised data - c. semistructured data - d. unstructured data **Recorded selection:** a — structured data --- ## Question 6 In order to know the correlation between one attribute to another attribute of data, we can use - a. Visualization and Pearson correlation - b. Visualization - c. Clustering - d. Pearson correlation **Recorded selection:** a — Visualization and Pearson correlation --- ## Question 7 A Supervised learning methods resulting explainable model is - a. linear regression - b. gradient descent algorithm - c. logistic regression - d. decision tree classifier **Recorded selection:** d — decision tree classifier --- ## Question 8 Line graph is suitable to present - a. image - b. video - c. tabular data - d. sequence data **Recorded selection:** c — tabular data --- ## Question 9 Please pay attention to the following confusion matrix, please calculate the Precision for class mango - a. 17% - b. 33% - c. 29% - d. 27% **Recorded selection:** *none — this question was left unanswered in the attempt.* > ⚠️ **Source defect:** the confusion matrix referenced by this question **does not appear** in the exported page — the image is missing from the PDF. The question is unanswerable from this source alone. It is almost certainly the same 3×3 matrix that appears in Q11 and Q13 (see below), which is reproduced here for completeness. --- ## Question 10 Gradient descent algorithm is an algorithm to find - a. error value - b. prediction value - c. appropriate values of coefficients in a linear function - d. label **Recorded selection:** c — appropriate values of coefficients in a linear function --- ## Question 11 Please pay attention to the following confusion matrix, please calculate the Overall Accuracy **Confusion matrix (shared by Q11 and Q13):** | | | **True Class** | | | |---|---|---|---|---| | | | Apple | Orange | Mango | | **Predicted Class** | Apple | 7 | 8 | 9 | | | Orange | 1 | 2 | 3 | | | Mango | 3 | 2 | 1 | - a. 33.98% - b. 27.99% - c. 17.03% - d. 47.10% **Recorded selection:** b — 27.99% > *Worked check:* diagonal sum = 7 + 2 + 1 = 10; total = 7+8+9+1+2+3+3+2+1 = 36. Accuracy = 10/36 = **27.78 %**. The nearest option is b (27.99 %), but the two do **not** agree exactly — a 0.21 pp gap that is larger than rounding, so the option was likely generated from a slightly different matrix or an unstated rounding step. Flagged rather than reconciled. --- ## Question 13 Please pay attention to the following confusion matrix, please calculate the F1-measure *(Uses the same matrix as Question 11.)* - a. 0.425 - b. 0.625 - c. 0.725 - d. 0.525 **Recorded selection:** b — 0.625 > *Worked check:* the option **cannot be reproduced** from the matrix in Q11 under any standard definition. Macro-averaged F1 (Apple 0.636, Orange 0.154, Mango 0.083) = **0.291**; micro-F1 (= accuracy) = **0.278**; neither is close to 0.625. Either the option set was computed from a different matrix, or "F1" here refers to a binary/other-class formulation not derivable from the figure. Flagged as inconsistent. > **Numbering note:** the quiz jumps **Q11 → Q13**. Question 12 is absent from the source PDF (confirmed at page level: page 11 = Q11, page 12 = Q13). It was either deleted from the quiz or dropped by the export. --- ## Question 14 In machine learning, cost function is same with - a. log function - b. square error function - c. absolute error function - d. error function **Recorded selection:** c — absolute error function --- ## Question 15 Please pay attention to the following confusion matrix, please calculate the Precision **Confusion matrix (shared by Q15 and Q16):** | | | **Real Label** | | |---|---|---|---| | | | Kelas 1 (Positif) | Kelas 2 (Negative) | | **Prediction** | Kelas 1 (Positif) | 25 | 14 | | | Kelas 2 (Negative) | 16 | 36 | - a. 60.98% - b. 69.03% - c. 67.03% - d. 64.10% **Recorded selection:** d — 64.10% > *Worked check:* Precision = 25 / (25 + 14) = 25/39 = 64.10 %. --- ## Question 16 Please pay attention to the following confusion matrix, please calculate the Accuracy *(Uses the same matrix as Question 15.)* - a. 67.03% - b. 64.10% - c. 69.03% - d. 60.98% **Recorded selection:** a — 67.03% > *Worked check:* Accuracy = (25 + 36) / (25+14+16+36) = 61/91 = 67.03 %. --- ## Question 17 There are several strategies to find suitable parameter for machine learning model, except - a. best guess - b. one factor at a time - c. grid searching - d. bruto force searching *(sic — source reads "bruto")* **Recorded selection:** d — bruto force searching --- ## Question 18 There are several ways for data acquisition from open data repositories, except - a. using data warehouse - b. using API - c. data acquisition manually - d. using web scraping **Recorded selection:** c — data acquisition manually --- ## Question 19 There are tasks which can be solved by supervised learning, except - a. anomaly detection - b. object detection - c. text classification - d. image captioning **Recorded selection:** a — anomaly detection --- ## Question 20 In Tukey's fences method, we can detect outlier using - a. third quartile - b. first quartile - c. second quartile *(sic — source reads "quatile")* - d. interquatile range *(sic — source reads "interquatile")* **Recorded selection:** d — interquatile range --- ## Question 21 When you have 1.000.000 data, the first step to develop machine learning model related the data is - a. Split the data to validation and testing data - b. Split the data to be training, validation and testing data - c. conducting k-fold cross validation - d. conducting stratified k-fold cross validation **Recorded selection:** b — Split the data to be training, validation and testing data --- ## Question 22 What type of visualization which is suitable to present 2 attributes/columns - a. line graph - b. scattet plot *(sic — source reads "scattet")* - c. histogram - d. heatmap **Recorded selection:** d — heatmap --- ## Question 23 After generate clusters using unsupervised learning algorithm, the next step is - a. deep learning - b. writing report - c. classification - d. interpretation **Recorded selection:** d — interpretation --- ## Transcription notes - **Source:** `Pembelajaran Mesin - UTS - 2022 - REG B(1).pdf` — 23 pages, **image-only scan with no text layer** (0 extractable characters); every page was rendered with `pypdfium2` at 2× and read visually. - **Nature of the document:** this is a **Moodle quiz attempt export**, i.e. screenshots of the quiz-taking interface, not a scanned exam paper. Every page therefore repeats the UI furniture *"Question N / Marked out of 1.00 / Flag question / Clear my choice"*, which is **not** reproduced here. Only the question stem, the options and the recorded selection are transcribed. - **"Recorded selection"** reflects the filled radio button in the export — i.e. the answer the student submitted — **not** a verified correct answer key. Two questions carry a visible *"Not yet answered"* / *"Remove flag"* state difference (Q9 unanswered; Q13, Q21 flagged). - **Question 12 is absent from the source.** Page 11 ends at Q11 and page 12 begins at Q13; there is no page containing Q12. The numbering gap is genuine and not a transcription omission. - **Page 23 is blank.** - **Two questions share one figure.** The 3×3 Apple/Orange/Mango matrix belongs to Q11 (accuracy) and Q13 (F1); the 2×2 Kelas 1/Kelas 2 matrix belongs to Q15 (precision) and Q16 (accuracy). The exporter did not re-render the figure on the second question of each pair, so the duplicate pages show only the option list — the matrix is reproduced once here and cross-referenced. - **Q9's figure is missing entirely** from the export; see the defect note at Q9. - **Spelling is preserved as printed**, including the source's own typos: *"distance-base"* (Q3), *"bruto force"* (Q17), *"quatile"* / *"interquatile"* (Q20), *"scattet plot"* (Q22). These are marked *(sic)* rather than silently corrected. - The worked arithmetic checks added under Q11, Q13, Q15 and Q16 are **not part of the source** — they are included to verify that the recorded selections are numerically consistent with the given matrices. - The Kelas 1/Kelas 2 matrix is labelled in Indonesian while the question text is in English; this mixed-language presentation is as it appears in the source.